{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# AMULETY: using embeddings for Machine Learning\n", "\n", "This tutorial demonstrates how to use AMULETY for downstream machine learning tasks with B-cell receptor (BCR) data. AMULETY is a python package for generating embeddings from BCR and TCR sequences using various state-of-the-art models.\n", "\n", "## Overview\n", "\n", "In this tutorial, you will learn to:\n", "1. Install and set up AMULETY\n", "2. Load and prepare BCR data in AIRR format\n", "3. Generate embeddings using different models\n", "4. Perform downstream machine learning tasks\n", "5. Evaluate model performance\n", "\n", "## Installation\n", "\n", "First, install AMULETY using pip:" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Requirement already satisfied: amulety in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (0.1.1)\n", "Requirement already satisfied: numpy in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (1.26.0)\n", "Requirement already satisfied: pandas in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (2.2.3)\n", "Requirement already satisfied: torch in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (2.6.0)\n", "Requirement already satisfied: transformers in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (4.50.3)\n", "Requirement already satisfied: typer in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (0.15.4)\n", "Requirement already satisfied: antiberty in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (0.1.3)\n", "Requirement already satisfied: ablang in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (0.3.1)\n", "Requirement already satisfied: rjieba in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (0.1.13)\n", "Requirement already satisfied: pre-commit in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (4.2.0)\n", "Requirement already satisfied: protobuf in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (4.25.6)\n", "Requirement already satisfied: sentencepiece in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (0.2.0)\n", "Requirement already satisfied: pytest-workflow>=1.6.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (2.1.0)\n", "Requirement already satisfied: pytest>=7.0.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (8.3.5)\n", "Requirement already satisfied: gensim>=3.8.3 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from amulety) (4.3.3)\n", "Requirement already satisfied: scipy<1.14.0,>=1.7.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from gensim>=3.8.3->amulety) (1.13.1)\n", "Requirement already satisfied: smart-open>=1.8.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from gensim>=3.8.3->amulety) (7.1.0)\n", "Requirement already satisfied: exceptiongroup>=1.0.0rc8 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest>=7.0.0->amulety) (1.2.0)\n", "Requirement already satisfied: iniconfig in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest>=7.0.0->amulety) (2.1.0)\n", "Requirement already satisfied: packaging in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest>=7.0.0->amulety) (24.2)\n", "Requirement already satisfied: pluggy<2,>=1.5 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest>=7.0.0->amulety) (1.6.0)\n", "Requirement already satisfied: tomli>=1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest>=7.0.0->amulety) (2.2.1)\n", "Requirement already satisfied: pyyaml in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest-workflow>=1.6.0->amulety) (6.0.2)\n", "Requirement already satisfied: jsonschema in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest-workflow>=1.6.0->amulety) (4.23.0)\n", "Requirement already satisfied: xopen>=1.4.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest-workflow>=1.6.0->amulety) (2.0.2)\n", "Requirement already satisfied: zstandard in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pytest-workflow>=1.6.0->amulety) (0.23.0)\n", "Requirement already satisfied: wrapt in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from smart-open>=1.8.1->gensim>=3.8.3->amulety) (1.14.1)\n", "Requirement already satisfied: requests in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from ablang->amulety) (2.32.4)\n", "Requirement already satisfied: numba in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from ablang->amulety) (0.61.0)\n", "Requirement already satisfied: filelock in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (3.17.0)\n", "Requirement already satisfied: typing-extensions>=4.10.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (4.12.2)\n", "Requirement already satisfied: networkx in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (3.4.2)\n", "Requirement already satisfied: jinja2 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (3.1.5)\n", "Requirement already satisfied: fsspec in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (2025.2.0)\n", "Requirement already satisfied: sympy==1.13.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch->amulety) (1.13.1)\n", "Requirement already satisfied: mpmath<1.4,>=1.1.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from sympy==1.13.1->torch->amulety) (1.3.0)\n", "Requirement already satisfied: huggingface-hub<1.0,>=0.26.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from transformers->amulety) (0.30.1)\n", "Requirement already satisfied: regex!=2019.12.17 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from transformers->amulety) (2024.11.6)\n", "Requirement already satisfied: tokenizers<0.22,>=0.21 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from transformers->amulety) (0.21.1)\n", "Requirement already satisfied: safetensors>=0.4.3 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from transformers->amulety) (0.5.3)\n", "Requirement already satisfied: tqdm>=4.27 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from transformers->amulety) (4.67.1)\n", "Requirement already satisfied: MarkupSafe>=2.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jinja2->torch->amulety) (3.0.2)\n", "Requirement already satisfied: attrs>=22.2.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jsonschema->pytest-workflow>=1.6.0->amulety) (24.3.0)\n", "Requirement already satisfied: jsonschema-specifications>=2023.03.6 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jsonschema->pytest-workflow>=1.6.0->amulety) (2024.10.1)\n", "Requirement already satisfied: referencing>=0.28.4 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jsonschema->pytest-workflow>=1.6.0->amulety) (0.36.2)\n", "Requirement already satisfied: rpds-py>=0.7.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jsonschema->pytest-workflow>=1.6.0->amulety) (0.23.1)\n", "Requirement already satisfied: llvmlite<0.45,>=0.44.0dev0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from numba->ablang->amulety) (0.44.0)\n", "Requirement already satisfied: python-dateutil>=2.8.2 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas->amulety) (2.9.0.post0)\n", "Requirement already satisfied: pytz>=2020.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas->amulety) (2025.1)\n", "Requirement already satisfied: tzdata>=2022.7 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas->amulety) (2025.1)\n", "Requirement already satisfied: six>=1.5 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from python-dateutil>=2.8.2->pandas->amulety) (1.17.0)\n", "Requirement already satisfied: cfgv>=2.0.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pre-commit->amulety) (3.4.0)\n", "Requirement already satisfied: identify>=1.0.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pre-commit->amulety) (2.6.12)\n", "Requirement already satisfied: nodeenv>=0.11.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pre-commit->amulety) (1.9.1)\n", "Requirement already satisfied: virtualenv>=20.10.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pre-commit->amulety) (20.31.2)\n", "Requirement already satisfied: distlib<1,>=0.3.7 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from virtualenv>=20.10.0->pre-commit->amulety) (0.3.9)\n", "Requirement already satisfied: platformdirs<5,>=3.9.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from virtualenv>=20.10.0->pre-commit->amulety) (4.3.7)\n", "Requirement already satisfied: charset_normalizer<4,>=2 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from requests->ablang->amulety) (3.3.2)\n", "Requirement already satisfied: idna<4,>=2.5 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from requests->ablang->amulety) (3.10)\n", "Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from requests->ablang->amulety) (1.26.20)\n", "Requirement already satisfied: certifi>=2017.4.17 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from requests->ablang->amulety) (2025.6.15)\n", "Requirement already satisfied: click<8.2,>=8.0.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from typer->amulety) (8.1.8)\n", "Requirement already satisfied: shellingham>=1.3.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from typer->amulety) (1.5.4)\n", "Requirement already satisfied: rich>=10.11.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from typer->amulety) (13.9.4)\n", "Requirement already satisfied: markdown-it-py>=2.2.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from rich>=10.11.0->typer->amulety) (3.0.0)\n", "Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from rich>=10.11.0->typer->amulety) (2.19.1)\n", "Requirement already satisfied: mdurl~=0.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from markdown-it-py>=2.2.0->rich>=10.11.0->typer->amulety) (0.1.2)\n", "Requirement already satisfied: pandas in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (2.2.3)\n", "Requirement already satisfied: numpy in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (1.26.0)\n", "Requirement already satisfied: scikit-learn in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (1.6.1)\n", "Requirement already satisfied: torch in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (2.6.0)\n", "Requirement already satisfied: matplotlib in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (3.10.0)\n", "Requirement already satisfied: seaborn in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (0.13.2)\n", "Requirement already satisfied: python-dateutil>=2.8.2 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas) (2.9.0.post0)\n", "Requirement already satisfied: pytz>=2020.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas) (2025.1)\n", "Requirement already satisfied: tzdata>=2022.7 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from pandas) (2025.1)\n", "Requirement already satisfied: scipy>=1.6.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from scikit-learn) (1.13.1)\n", "Requirement already satisfied: joblib>=1.2.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from scikit-learn) (1.4.2)\n", "Requirement already satisfied: threadpoolctl>=3.1.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from scikit-learn) (3.5.0)\n", "Requirement already satisfied: filelock in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (3.17.0)\n", "Requirement already satisfied: typing-extensions>=4.10.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (4.12.2)\n", "Requirement already satisfied: networkx in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (3.4.2)\n", "Requirement already satisfied: jinja2 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (3.1.5)\n", "Requirement already satisfied: fsspec in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (2025.2.0)\n", "Requirement already satisfied: sympy==1.13.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from torch) (1.13.1)\n", "Requirement already satisfied: mpmath<1.4,>=1.1.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from sympy==1.13.1->torch) (1.3.0)\n", "Requirement already satisfied: contourpy>=1.0.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (1.3.1)\n", "Requirement already satisfied: cycler>=0.10 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (0.11.0)\n", "Requirement already satisfied: fonttools>=4.22.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (4.55.3)\n", "Requirement already satisfied: kiwisolver>=1.3.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (1.4.8)\n", "Requirement already satisfied: packaging>=20.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (24.2)\n", "Requirement already satisfied: pillow>=8 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (11.1.0)\n", "Requirement already satisfied: pyparsing>=2.3.1 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from matplotlib) (3.2.0)\n", "Requirement already satisfied: six>=1.5 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from python-dateutil>=2.8.2->pandas) (1.17.0)\n", "Requirement already satisfied: MarkupSafe>=2.0 in /opt/anaconda3/envs/torchen/lib/python3.10/site-packages (from jinja2->torch) (3.0.2)\n" ] } ], "source": [ "# Install Amulety\n", "!pip install amulety\n", "\n", "# Install additional dependencies for this tutorial\n", "!pip install pandas numpy scikit-learn torch matplotlib seaborn" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Import Required Libraries" ] }, { "cell_type": "code", "execution_count": 59, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "All libraries imported successfully!\n" ] } ], "source": [ "import pandas as pd\n", "import numpy as np\n", "import torch\n", "import matplotlib.pyplot as plt\n", "import seaborn as sns\n", "from sklearn.model_selection import StratifiedGroupKFold, GridSearchCV\n", "from sklearn.svm import SVC\n", "from sklearn.metrics import matthews_corrcoef, f1_score, accuracy_score\n", "from sklearn.preprocessing import LabelEncoder\n", "from collections import Counter\n", "import warnings\n", "warnings.filterwarnings('ignore')\n", "\n", "# Import Amulety\n", "from amulety import embed_airr\n", "\n", "print(\"All libraries imported successfully!\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Load BCR Data\n", "\n", "We'll load the BCR data in AIRR (Adaptive Immune Receptor Repertoire) format. This dataset contains antibody sequences with associated metadata including gene usage, isotype, and other annotations." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "--2025-09-24 13:32:27-- https://zenodo.org/records/17186858/files/ML_bcr_airr_dataset.tsv\n", "Resolving zenodo.org (zenodo.org)... 188.185.43.25, 188.185.45.92, 188.185.48.194, ...\n", "Connecting to zenodo.org (zenodo.org)|188.185.43.25|:443... connected.\n", "HTTP request sent, awaiting response... 200 OK\n", "Length: 345071483 (329M) [application/octet-stream]\n", "Saving to: 'tutorial/ML_bcr_airr_dataset.tsv'\n", "\n", "ML_bcr_airr_dataset 100%[===================>] 329.08M 6.45MB/s in 54s \n", "\n", "2025-09-24 13:33:21 (6.14 MB/s) - 'tutorial/ML_bcr_airr_dataset.tsv' saved [345071483/345071483]\n", "\n" ] } ], "source": [ "# Download the tutorial dataset\n", "! mkdir -p tutorial\n", "! wget -P tutorial https://zenodo.org/records/17186858/files/ML_bcr_airr_dataset.tsv" ] }, { "cell_type": "code", "execution_count": 60, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Dataset shape: (1399366, 19)\n", "\n", "Columns: ['sequence_id', 'sequence_vdj_aa', 'locus', 'cell_id', 'chain_type', 'v_call', 'v_call_family', 'j_call_family', 'mu_freq', 'junction_aa_length', 'isotype', 'source', 'subject', 'specificity', 'duplicate_count', 'productive', 'rev_comp', 'stop_codon', 'vj_in_frame']\n", "\n", "First few rows:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
sequence_idsequence_vdj_aalocuscell_idchain_typev_callv_call_familyj_call_familymu_freqjunction_aa_lengthisotypesourcesubjectspecificityduplicate_countproductiverev_compstop_codonvj_in_frame
01_heavyQVQLVESGGGLVKPGGSLRLSCAASGFTFSDYYMSWIRQAPGKGLE...IGHcell_1HIGHV3-11IGHV3IGHJ40.01644712.0NaNOASOAS_King_Subject-BCP3unlabeled1TrueFalseFalseTrue
12_heavyQVQLVESGGGVVQPGRSLRLSCAASGFTFSSYAMHWVRQAPGKGLE...IGHcell_2HIGHV3-30IGHV3IGHJ40.02657817.0NaNOASOAS_King_Subject-BCP3unlabeled1TrueFalseFalseTrue
23_heavyEVQLVESGGGLVKPGGSLTLSCAVSGFTFKNAWMSWVRQAPGKGLE...IGHcell_3HIGHV3-15IGHV3IGHJ40.06953614.0NaNOASOAS_King_Subject-BCP3unlabeled1TrueFalseFalseTrue
34_heavyEVQLVESGGALVKPGGSLRLSCVVSGLTFTDAYMIWVRQAPGKGLE...IGHcell_4HIGHV3-15IGHV3IGHJ60.07541014.0NaNOASOAS_King_Subject-BCP3unlabeled1TrueFalseFalseTrue
45_heavyQEELVEAGGTVVQPGRSLGLSCAASGFSFSNYLMHWVRQTPGKGLE...IGHcell_5HIGHV3-30-3IGHV3IGHJ50.07692322.0NaNOASOAS_King_Subject-BCP3unlabeled1TrueFalseFalseTrue
\n", "
" ], "text/plain": [ " sequence_id sequence_vdj_aa locus \\\n", "0 1_heavy QVQLVESGGGLVKPGGSLRLSCAASGFTFSDYYMSWIRQAPGKGLE... IGH \n", "1 2_heavy QVQLVESGGGVVQPGRSLRLSCAASGFTFSSYAMHWVRQAPGKGLE... IGH \n", "2 3_heavy EVQLVESGGGLVKPGGSLTLSCAVSGFTFKNAWMSWVRQAPGKGLE... IGH \n", "3 4_heavy EVQLVESGGALVKPGGSLRLSCVVSGLTFTDAYMIWVRQAPGKGLE... IGH \n", "4 5_heavy QEELVEAGGTVVQPGRSLGLSCAASGFSFSNYLMHWVRQTPGKGLE... IGH \n", "\n", " cell_id chain_type v_call v_call_family j_call_family mu_freq \\\n", "0 cell_1 H IGHV3-11 IGHV3 IGHJ4 0.016447 \n", "1 cell_2 H IGHV3-30 IGHV3 IGHJ4 0.026578 \n", "2 cell_3 H IGHV3-15 IGHV3 IGHJ4 0.069536 \n", "3 cell_4 H IGHV3-15 IGHV3 IGHJ6 0.075410 \n", "4 cell_5 H IGHV3-30-3 IGHV3 IGHJ5 0.076923 \n", "\n", " junction_aa_length isotype source subject specificity \\\n", "0 12.0 NaN OAS OAS_King_Subject-BCP3 unlabeled \n", "1 17.0 NaN OAS OAS_King_Subject-BCP3 unlabeled \n", "2 14.0 NaN OAS OAS_King_Subject-BCP3 unlabeled \n", "3 14.0 NaN OAS OAS_King_Subject-BCP3 unlabeled \n", "4 22.0 NaN OAS OAS_King_Subject-BCP3 unlabeled \n", "\n", " duplicate_count productive rev_comp stop_codon vj_in_frame \n", "0 1 True False False True \n", "1 1 True False False True \n", "2 1 True False False True \n", "3 1 True False False True \n", "4 1 True False False True " ] }, "execution_count": 60, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# Load the BCR AIRR dataset\n", "bcr_data = pd.read_csv(\"tutorial/ML_bcr_airr_dataset.tsv\", sep='\\t')\n", "\n", "print(f\"Dataset shape: {bcr_data.shape}\")\n", "print(f\"\\nColumns: {list(bcr_data.columns)}\")\n", "print(f\"\\nFirst few rows:\")\n", "bcr_data.head()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Data Exploration\n", "\n", "Let's explore the dataset to understand its structure and characteristics." ] }, { "cell_type": "code", "execution_count": 61, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Dataset Statistics:\n", "Total sequences: 1399366\n", "Unique subjects: 74\n", "Chain types: {'H': 858682, 'L': 540684}\n", "Locus distribution: {'IGH': 858682, 'IGK': 314772, 'IGL': 225912}\n", "\n", "Sequence length statistics:\n", "count 1.399366e+06\n", "mean 1.171950e+02\n", "std 8.253170e+00\n", "min 2.100000e+01\n", "25% 1.100000e+02\n", "50% 1.190000e+02\n", "75% 1.230000e+02\n", "max 1.860000e+02\n", "Name: sequence_length, dtype: float64\n" ] } ], "source": [ "# Basic statistics\n", "print(\"Dataset Statistics:\")\n", "print(f\"Total sequences: {len(bcr_data)}\")\n", "print(f\"Unique subjects: {bcr_data['subject'].nunique()}\")\n", "print(f\"Chain types: {bcr_data['chain_type'].value_counts().to_dict()}\")\n", "print(f\"Locus distribution: {bcr_data['locus'].value_counts().to_dict()}\")\n", "\n", "# Check sequence length distribution\n", "bcr_data['sequence_length'] = bcr_data['sequence_vdj_aa'].str.len()\n", "print(f\"\\nSequence length statistics:\")\n", "print(bcr_data['sequence_length'].describe())" ] }, { "cell_type": "code", "execution_count": 62, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Visualize sequence length distribution\n", "plt.figure(figsize=(12, 4))\n", "\n", "plt.subplot(1, 2, 1)\n", "plt.hist(bcr_data['sequence_length'], bins=50, alpha=0.7, edgecolor='black')\n", "plt.xlabel('Sequence Length')\n", "plt.ylabel('Frequency')\n", "plt.title('Distribution of Sequence Lengths')\n", "\n", "plt.subplot(1, 2, 2)\n", "chain_counts = bcr_data['chain_type'].value_counts()\n", "plt.pie(chain_counts.values, labels=chain_counts.index, autopct='%1.1f%%')\n", "plt.title('Chain Type Distribution')\n", "\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Generate Embeddings with Amulety\n", "\n", "Now we'll use Amulety to generate embeddings from our BCR sequences. Amulety supports multiple embedding models including BCR-specific models and protein language models." ] }, { "cell_type": "code", "execution_count": 63, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Using 858682 heavy chain sequences for embedding\n", "Working with 10000 sequences for demonstration\n" ] } ], "source": [ "# Prepare data for embedding\n", "# Filter for heavy chains only for this example\n", "heavy_chains = bcr_data[bcr_data['chain_type'] == 'H'].copy()\n", "print(f\"Using {len(heavy_chains)} heavy chain sequences for embedding\")\n", "\n", "# Take a subset for demonstration \n", "# please adjust size based on computational resources)\n", "sample_size = min(10000, len(heavy_chains))\n", "heavy_sample = heavy_chains.sample(n=sample_size, random_state=42).reset_index(drop=True)\n", "print(f\"Working with {len(heavy_sample)} sequences for demonstration\")" ] }, { "cell_type": "code", "execution_count": 64, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2025-09-24 13:37:22,906 - INFO - Detected single-cell data format\n", "2025-09-24 13:37:22,912 - INFO - Single-cell AIRR data detected (all entries have cell_id).\n", "2025-09-24 13:37:23,035 - INFO - AntiBERTy loaded. Size: 26.03 M\n", "2025-09-24 13:37:23,035 - INFO - Batch 1/1250\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Generating embeddings using antiBERTy model...\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "2025-09-24 13:37:23,122 - INFO - Batch 2/1250\n", "2025-09-24 13:37:23,188 - INFO - Batch 3/1250\n", "2025-09-24 13:37:23,244 - INFO - Batch 4/1250\n", "2025-09-24 13:37:23,306 - INFO - Batch 5/1250\n", "2025-09-24 13:37:23,366 - INFO - Batch 6/1250\n", "2025-09-24 13:37:23,425 - INFO - Batch 7/1250\n", "2025-09-24 13:37:23,491 - INFO - Batch 8/1250\n", "2025-09-24 13:37:23,579 - INFO - Batch 9/1250\n", "2025-09-24 13:37:23,642 - INFO - Batch 10/1250\n", "2025-09-24 13:37:23,716 - INFO - Batch 11/1250\n", "2025-09-24 13:37:23,774 - INFO - Batch 12/1250\n", "2025-09-24 13:37:23,835 - INFO - Batch 13/1250\n", "2025-09-24 13:37:23,893 - INFO - Batch 14/1250\n", "2025-09-24 13:37:23,958 - INFO - Batch 15/1250\n", "2025-09-24 13:37:24,024 - INFO - Batch 16/1250\n", "2025-09-24 13:37:24,088 - INFO - Batch 17/1250\n", "2025-09-24 13:37:24,154 - INFO - Batch 18/1250\n", "2025-09-24 13:37:24,213 - INFO - Batch 19/1250\n", "2025-09-24 13:37:24,292 - INFO - Batch 20/1250\n", "2025-09-24 13:37:24,358 - INFO - Batch 21/1250\n", "2025-09-24 13:37:24,424 - INFO - Batch 22/1250\n", "2025-09-24 13:37:24,488 - INFO - Batch 23/1250\n", "2025-09-24 13:37:24,553 - INFO - Batch 24/1250\n", "2025-09-24 13:37:24,612 - INFO - Batch 25/1250\n", "2025-09-24 13:37:24,677 - INFO - Batch 26/1250\n", "2025-09-24 13:37:24,747 - INFO - Batch 27/1250\n", "2025-09-24 13:37:24,808 - INFO - Batch 28/1250\n", "2025-09-24 13:37:24,870 - INFO - Batch 29/1250\n", "2025-09-24 13:37:24,937 - INFO - Batch 30/1250\n", "2025-09-24 13:37:25,005 - INFO - Batch 31/1250\n", "2025-09-24 13:37:25,069 - INFO - Batch 32/1250\n", "2025-09-24 13:37:25,127 - INFO - Batch 33/1250\n", "2025-09-24 13:37:25,197 - INFO - Batch 34/1250\n", "2025-09-24 13:37:25,264 - INFO - Batch 35/1250\n", "2025-09-24 13:37:25,333 - INFO - Batch 36/1250\n", "2025-09-24 13:37:25,401 - INFO - Batch 37/1250\n", "2025-09-24 13:37:25,469 - INFO - Batch 38/1250\n", "2025-09-24 13:37:25,536 - INFO - Batch 39/1250\n", "2025-09-24 13:37:25,603 - INFO - Batch 40/1250\n", "2025-09-24 13:37:25,669 - INFO - Batch 41/1250\n", "2025-09-24 13:37:25,735 - INFO - Batch 42/1250\n", "2025-09-24 13:37:25,794 - INFO - Batch 43/1250\n", "2025-09-24 13:37:25,856 - INFO - Batch 44/1250\n", "2025-09-24 13:37:25,919 - INFO - Batch 45/1250\n", "2025-09-24 13:37:25,980 - INFO - Batch 46/1250\n", "2025-09-24 13:37:26,058 - INFO - Batch 47/1250\n", "2025-09-24 13:37:26,119 - INFO - Batch 48/1250\n", "2025-09-24 13:37:26,183 - INFO - Batch 49/1250\n", "2025-09-24 13:37:26,241 - INFO - Batch 50/1250\n", "2025-09-24 13:37:26,302 - INFO - Batch 51/1250\n", "2025-09-24 13:37:26,363 - INFO - Batch 52/1250\n", "2025-09-24 13:37:26,426 - INFO - Batch 53/1250\n", "2025-09-24 13:37:26,488 - INFO - Batch 54/1250\n", "2025-09-24 13:37:26,557 - INFO - Batch 55/1250\n", "2025-09-24 13:37:26,622 - INFO - Batch 56/1250\n", "2025-09-24 13:37:26,686 - INFO - Batch 57/1250\n", "2025-09-24 13:37:26,748 - INFO - Batch 58/1250\n", "2025-09-24 13:37:26,812 - INFO - Batch 59/1250\n", "2025-09-24 13:37:26,872 - INFO - Batch 60/1250\n", "2025-09-24 13:37:26,932 - INFO - Batch 61/1250\n", "2025-09-24 13:37:27,001 - INFO - Batch 62/1250\n", "2025-09-24 13:37:27,064 - INFO - Batch 63/1250\n", "2025-09-24 13:37:27,127 - INFO - Batch 64/1250\n", "2025-09-24 13:37:27,186 - INFO - Batch 65/1250\n", "2025-09-24 13:37:27,250 - INFO - Batch 66/1250\n", "2025-09-24 13:37:27,316 - INFO - Batch 67/1250\n", "2025-09-24 13:37:27,377 - INFO - Batch 68/1250\n", "2025-09-24 13:37:27,438 - INFO - Batch 69/1250\n", "2025-09-24 13:37:27,501 - INFO - Batch 70/1250\n", "2025-09-24 13:37:27,567 - INFO - Batch 71/1250\n", "2025-09-24 13:37:27,632 - INFO - Batch 72/1250\n", "2025-09-24 13:37:27,697 - INFO - Batch 73/1250\n", "2025-09-24 13:37:27,761 - INFO - Batch 74/1250\n", "2025-09-24 13:37:27,826 - INFO - Batch 75/1250\n", "2025-09-24 13:37:27,897 - INFO - Batch 76/1250\n", "2025-09-24 13:37:27,966 - INFO - Batch 77/1250\n", "2025-09-24 13:37:28,028 - INFO - Batch 78/1250\n", "2025-09-24 13:37:28,092 - INFO - Batch 79/1250\n", "2025-09-24 13:37:28,153 - INFO - Batch 80/1250\n", "2025-09-24 13:37:28,222 - INFO - Batch 81/1250\n", "2025-09-24 13:37:28,284 - INFO - Batch 82/1250\n", "2025-09-24 13:37:28,345 - INFO - Batch 83/1250\n", "2025-09-24 13:37:28,474 - INFO - Batch 84/1250\n", "2025-09-24 13:37:28,536 - INFO - Batch 85/1250\n", "2025-09-24 13:37:28,610 - INFO - Batch 86/1250\n", "2025-09-24 13:37:28,675 - INFO - Batch 87/1250\n", "2025-09-24 13:37:28,744 - INFO - Batch 88/1250\n", "2025-09-24 13:37:28,811 - INFO - Batch 89/1250\n", "2025-09-24 13:37:28,875 - INFO - Batch 90/1250\n", "2025-09-24 13:37:28,944 - INFO - Batch 91/1250\n", "2025-09-24 13:37:29,005 - INFO - Batch 92/1250\n", "2025-09-24 13:37:29,066 - INFO - Batch 93/1250\n", "2025-09-24 13:37:29,131 - INFO - Batch 94/1250\n", "2025-09-24 13:37:29,197 - INFO - Batch 95/1250\n", "2025-09-24 13:37:29,263 - INFO - Batch 96/1250\n", "2025-09-24 13:37:29,326 - INFO - Batch 97/1250\n", "2025-09-24 13:37:29,387 - INFO - Batch 98/1250\n", "2025-09-24 13:37:29,453 - INFO - Batch 99/1250\n", "2025-09-24 13:37:29,530 - INFO - Batch 100/1250\n", "2025-09-24 13:37:29,596 - INFO - Batch 101/1250\n", "2025-09-24 13:37:29,662 - INFO - Batch 102/1250\n", "2025-09-24 13:37:29,727 - INFO - Batch 103/1250\n", "2025-09-24 13:37:29,795 - INFO - Batch 104/1250\n", "2025-09-24 13:37:29,860 - INFO - Batch 105/1250\n", "2025-09-24 13:37:29,923 - INFO - Batch 106/1250\n", "2025-09-24 13:37:29,987 - INFO - Batch 107/1250\n", "2025-09-24 13:37:30,049 - INFO - Batch 108/1250\n", "2025-09-24 13:37:30,113 - INFO - Batch 109/1250\n", "2025-09-24 13:37:30,175 - INFO - Batch 110/1250\n", "2025-09-24 13:37:30,237 - INFO - Batch 111/1250\n", "2025-09-24 13:37:30,303 - INFO - Batch 112/1250\n", "2025-09-24 13:37:30,366 - INFO - Batch 113/1250\n", "2025-09-24 13:37:30,432 - INFO - Batch 114/1250\n", "2025-09-24 13:37:30,500 - INFO - Batch 115/1250\n", "2025-09-24 13:37:30,604 - INFO - Batch 116/1250\n", "2025-09-24 13:37:30,698 - INFO - Batch 117/1250\n", "2025-09-24 13:37:30,803 - INFO - Batch 118/1250\n", "2025-09-24 13:37:30,864 - INFO - Batch 119/1250\n", "2025-09-24 13:37:30,938 - INFO - Batch 120/1250\n", "2025-09-24 13:37:31,011 - INFO - Batch 121/1250\n", "2025-09-24 13:37:31,116 - INFO - Batch 122/1250\n", "2025-09-24 13:37:31,179 - INFO - Batch 123/1250\n", "2025-09-24 13:37:31,304 - INFO - Batch 124/1250\n", "2025-09-24 13:37:31,367 - INFO - Batch 125/1250\n", "2025-09-24 13:37:31,437 - INFO - Batch 126/1250\n", "2025-09-24 13:37:31,496 - INFO - Batch 127/1250\n", "2025-09-24 13:37:31,561 - INFO - Batch 128/1250\n", "2025-09-24 13:37:31,637 - INFO - Batch 129/1250\n", "2025-09-24 13:37:31,712 - INFO - Batch 130/1250\n", "2025-09-24 13:37:31,786 - INFO - Batch 131/1250\n", "2025-09-24 13:37:31,858 - INFO - Batch 132/1250\n", "2025-09-24 13:37:31,938 - INFO - Batch 133/1250\n", "2025-09-24 13:37:32,013 - INFO - Batch 134/1250\n", "2025-09-24 13:37:32,088 - INFO - Batch 135/1250\n", "2025-09-24 13:37:32,161 - INFO - Batch 136/1250\n", "2025-09-24 13:37:32,234 - INFO - Batch 137/1250\n", "2025-09-24 13:37:32,310 - INFO - Batch 138/1250\n", "2025-09-24 13:37:32,394 - INFO - Batch 139/1250\n", "2025-09-24 13:37:32,464 - INFO - Batch 140/1250\n", "2025-09-24 13:37:32,532 - INFO - Batch 141/1250\n", "2025-09-24 13:37:32,603 - INFO - Batch 142/1250\n", "2025-09-24 13:37:32,670 - INFO - Batch 143/1250\n", "2025-09-24 13:37:32,741 - INFO - Batch 144/1250\n", "2025-09-24 13:37:32,811 - INFO - Batch 145/1250\n", "2025-09-24 13:37:32,890 - INFO - Batch 146/1250\n", "2025-09-24 13:37:32,971 - INFO - Batch 147/1250\n", "2025-09-24 13:37:33,047 - INFO - Batch 148/1250\n", "2025-09-24 13:37:33,120 - INFO - Batch 149/1250\n", "2025-09-24 13:37:33,189 - INFO - Batch 150/1250\n", "2025-09-24 13:37:33,259 - INFO - Batch 151/1250\n", "2025-09-24 13:37:33,325 - INFO - Batch 152/1250\n", "2025-09-24 13:37:33,398 - INFO - Batch 153/1250\n", "2025-09-24 13:37:33,475 - INFO - Batch 154/1250\n", "2025-09-24 13:37:33,546 - INFO - Batch 155/1250\n", "2025-09-24 13:37:33,617 - INFO - Batch 156/1250\n", "2025-09-24 13:37:33,688 - INFO - Batch 157/1250\n", "2025-09-24 13:37:33,762 - INFO - Batch 158/1250\n", "2025-09-24 13:37:33,840 - INFO - Batch 159/1250\n", "2025-09-24 13:37:33,911 - INFO - Batch 160/1250\n", "2025-09-24 13:37:33,984 - INFO - Batch 161/1250\n", "2025-09-24 13:37:34,055 - INFO - Batch 162/1250\n", "2025-09-24 13:37:34,120 - INFO - Batch 163/1250\n", "2025-09-24 13:37:34,180 - INFO - Batch 164/1250\n", "2025-09-24 13:37:34,241 - INFO - Batch 165/1250\n", "2025-09-24 13:37:34,300 - INFO - Batch 166/1250\n", "2025-09-24 13:37:34,366 - INFO - Batch 167/1250\n", "2025-09-24 13:37:34,431 - INFO - Batch 168/1250\n", "2025-09-24 13:37:34,498 - INFO - Batch 169/1250\n", "2025-09-24 13:37:34,561 - INFO - Batch 170/1250\n", "2025-09-24 13:37:34,622 - INFO - Batch 171/1250\n", "2025-09-24 13:37:34,688 - INFO - Batch 172/1250\n", "2025-09-24 13:37:34,757 - INFO - Batch 173/1250\n", "2025-09-24 13:37:34,819 - INFO - Batch 174/1250\n", "2025-09-24 13:37:34,883 - INFO - Batch 175/1250\n", "2025-09-24 13:37:34,944 - INFO - Batch 176/1250\n", "2025-09-24 13:37:35,006 - INFO - Batch 177/1250\n", "2025-09-24 13:37:35,068 - INFO - Batch 178/1250\n", "2025-09-24 13:37:35,131 - INFO - Batch 179/1250\n", "2025-09-24 13:37:35,192 - INFO - Batch 180/1250\n", "2025-09-24 13:37:35,253 - INFO - Batch 181/1250\n", "2025-09-24 13:37:35,369 - INFO - Batch 182/1250\n", "2025-09-24 13:37:35,431 - INFO - Batch 183/1250\n", "2025-09-24 13:37:35,492 - INFO - Batch 184/1250\n", "2025-09-24 13:37:35,553 - INFO - Batch 185/1250\n", "2025-09-24 13:37:35,616 - INFO - Batch 186/1250\n", "2025-09-24 13:37:35,683 - INFO - Batch 187/1250\n", "2025-09-24 13:37:35,744 - INFO - Batch 188/1250\n", "2025-09-24 13:37:35,805 - INFO - Batch 189/1250\n", "2025-09-24 13:37:35,863 - INFO - Batch 190/1250\n", "2025-09-24 13:37:35,925 - INFO - Batch 191/1250\n", "2025-09-24 13:37:35,985 - INFO - Batch 192/1250\n", "2025-09-24 13:37:36,046 - INFO - Batch 193/1250\n", "2025-09-24 13:37:36,106 - INFO - Batch 194/1250\n", "2025-09-24 13:37:36,168 - INFO - Batch 195/1250\n", "2025-09-24 13:37:36,233 - INFO - Batch 196/1250\n", "2025-09-24 13:37:36,301 - INFO - Batch 197/1250\n", "2025-09-24 13:37:36,368 - INFO - Batch 198/1250\n", "2025-09-24 13:37:36,436 - INFO - Batch 199/1250\n", "2025-09-24 13:37:36,519 - INFO - Batch 200/1250\n", "2025-09-24 13:37:36,584 - INFO - Batch 201/1250\n", "2025-09-24 13:37:36,665 - INFO - Batch 202/1250\n", "2025-09-24 13:37:36,816 - INFO - Batch 203/1250\n", "2025-09-24 13:37:36,882 - INFO - Batch 204/1250\n", "2025-09-24 13:37:36,951 - INFO - Batch 205/1250\n", "2025-09-24 13:37:37,011 - INFO - Batch 206/1250\n", "2025-09-24 13:37:37,077 - INFO - Batch 207/1250\n", "2025-09-24 13:37:37,163 - INFO - Batch 208/1250\n", "2025-09-24 13:37:37,240 - INFO - Batch 209/1250\n", "2025-09-24 13:37:37,334 - INFO - Batch 210/1250\n", "2025-09-24 13:37:37,418 - INFO - Batch 211/1250\n", "2025-09-24 13:37:37,502 - INFO - Batch 212/1250\n", "2025-09-24 13:37:37,579 - INFO - Batch 213/1250\n", "2025-09-24 13:37:37,652 - INFO - Batch 214/1250\n", "2025-09-24 13:37:37,723 - INFO - Batch 215/1250\n", "2025-09-24 13:37:37,789 - INFO - Batch 216/1250\n", "2025-09-24 13:37:37,858 - INFO - Batch 217/1250\n", "2025-09-24 13:37:37,930 - INFO - Batch 218/1250\n", "2025-09-24 13:37:38,008 - INFO - Batch 219/1250\n", "2025-09-24 13:37:38,083 - INFO - Batch 220/1250\n", "2025-09-24 13:37:38,153 - INFO - Batch 221/1250\n", "2025-09-24 13:37:38,225 - INFO - Batch 222/1250\n", "2025-09-24 13:37:38,298 - INFO - Batch 223/1250\n", "2025-09-24 13:37:38,369 - INFO - Batch 224/1250\n", "2025-09-24 13:37:38,444 - INFO - Batch 225/1250\n", "2025-09-24 13:37:38,517 - INFO - Batch 226/1250\n", "2025-09-24 13:37:38,593 - INFO - Batch 227/1250\n", "2025-09-24 13:37:38,667 - INFO - Batch 228/1250\n", "2025-09-24 13:37:38,739 - INFO - Batch 229/1250\n", "2025-09-24 13:37:38,814 - INFO - Batch 230/1250\n", "2025-09-24 13:37:38,892 - INFO - Batch 231/1250\n", "2025-09-24 13:37:38,962 - INFO - Batch 232/1250\n", "2025-09-24 13:37:39,031 - INFO - Batch 233/1250\n", "2025-09-24 13:37:39,105 - INFO - Batch 234/1250\n", "2025-09-24 13:37:39,178 - INFO - Batch 235/1250\n", "2025-09-24 13:37:39,252 - INFO - Batch 236/1250\n", "2025-09-24 13:37:39,325 - INFO - Batch 237/1250\n", "2025-09-24 13:37:39,409 - INFO - Batch 238/1250\n", "2025-09-24 13:37:39,486 - INFO - Batch 239/1250\n", "2025-09-24 13:37:39,552 - INFO - Batch 240/1250\n", "2025-09-24 13:37:39,615 - INFO - Batch 241/1250\n", "2025-09-24 13:37:39,679 - INFO - Batch 242/1250\n", "2025-09-24 13:37:39,741 - INFO - Batch 243/1250\n", "2025-09-24 13:37:39,812 - INFO - Batch 244/1250\n", "2025-09-24 13:37:39,875 - INFO - Batch 245/1250\n", "2025-09-24 13:37:39,937 - INFO - Batch 246/1250\n", "2025-09-24 13:37:40,002 - INFO - Batch 247/1250\n", "2025-09-24 13:37:40,063 - INFO - Batch 248/1250\n", "2025-09-24 13:37:40,125 - INFO - Batch 249/1250\n", "2025-09-24 13:37:40,186 - INFO - Batch 250/1250\n", "2025-09-24 13:37:40,249 - INFO - Batch 251/1250\n", "2025-09-24 13:37:40,311 - INFO - Batch 252/1250\n", "2025-09-24 13:37:40,374 - INFO - Batch 253/1250\n", "2025-09-24 13:37:40,433 - INFO - Batch 254/1250\n", "2025-09-24 13:37:40,497 - INFO - Batch 255/1250\n", "2025-09-24 13:37:40,563 - INFO - Batch 256/1250\n", "2025-09-24 13:37:40,624 - INFO - Batch 257/1250\n", "2025-09-24 13:37:40,686 - INFO - Batch 258/1250\n", "2025-09-24 13:37:40,749 - INFO - Batch 259/1250\n", "2025-09-24 13:37:40,811 - INFO - Batch 260/1250\n", "2025-09-24 13:37:40,890 - INFO - Batch 261/1250\n", "2025-09-24 13:37:40,967 - INFO - Batch 262/1250\n", "2025-09-24 13:37:41,059 - INFO - Batch 263/1250\n", "2025-09-24 13:37:41,144 - INFO - Batch 264/1250\n", "2025-09-24 13:37:41,213 - INFO - Batch 265/1250\n", "2025-09-24 13:37:41,290 - INFO - Batch 266/1250\n", "2025-09-24 13:37:41,353 - INFO - Batch 267/1250\n", "2025-09-24 13:37:41,414 - INFO - Batch 268/1250\n", "2025-09-24 13:37:41,475 - INFO - Batch 269/1250\n", "2025-09-24 13:37:41,535 - INFO - Batch 270/1250\n", "2025-09-24 13:37:41,598 - INFO - Batch 271/1250\n", "2025-09-24 13:37:41,661 - INFO - Batch 272/1250\n", "2025-09-24 13:37:41,722 - INFO - Batch 273/1250\n", "2025-09-24 13:37:41,804 - INFO - Batch 274/1250\n", "2025-09-24 13:37:41,907 - INFO - Batch 275/1250\n", "2025-09-24 13:37:41,974 - INFO - Batch 276/1250\n", "2025-09-24 13:37:42,036 - INFO - Batch 277/1250\n", "2025-09-24 13:37:42,098 - INFO - Batch 278/1250\n", "2025-09-24 13:37:42,158 - INFO - Batch 279/1250\n", "2025-09-24 13:37:42,220 - INFO - Batch 280/1250\n", "2025-09-24 13:37:42,281 - INFO - Batch 281/1250\n", "2025-09-24 13:37:42,342 - INFO - Batch 282/1250\n", "2025-09-24 13:37:42,404 - INFO - Batch 283/1250\n", "2025-09-24 13:37:42,466 - INFO - Batch 284/1250\n", "2025-09-24 13:37:42,526 - INFO - Batch 285/1250\n", "2025-09-24 13:37:42,587 - INFO - Batch 286/1250\n", "2025-09-24 13:37:42,652 - INFO - Batch 287/1250\n", "2025-09-24 13:37:42,715 - INFO - Batch 288/1250\n", "2025-09-24 13:37:42,776 - INFO - Batch 289/1250\n", "2025-09-24 13:37:42,836 - INFO - Batch 290/1250\n", "2025-09-24 13:37:42,902 - INFO - Batch 291/1250\n", "2025-09-24 13:37:42,966 - INFO - Batch 292/1250\n", "2025-09-24 13:37:43,025 - INFO - Batch 293/1250\n", "2025-09-24 13:37:43,113 - INFO - Batch 294/1250\n", "2025-09-24 13:37:43,175 - INFO - Batch 295/1250\n", "2025-09-24 13:37:43,239 - INFO - Batch 296/1250\n", "2025-09-24 13:37:43,298 - INFO - Batch 297/1250\n", "2025-09-24 13:37:43,363 - INFO - Batch 298/1250\n", "2025-09-24 13:37:43,422 - INFO - Batch 299/1250\n", "2025-09-24 13:37:43,486 - INFO - Batch 300/1250\n", "2025-09-24 13:37:43,547 - INFO - Batch 301/1250\n", "2025-09-24 13:37:43,610 - INFO - Batch 302/1250\n", "2025-09-24 13:37:43,669 - INFO - Batch 303/1250\n", "2025-09-24 13:37:43,734 - INFO - Batch 304/1250\n", "2025-09-24 13:37:43,798 - INFO - Batch 305/1250\n", "2025-09-24 13:37:43,862 - INFO - Batch 306/1250\n", "2025-09-24 13:37:43,921 - INFO - Batch 307/1250\n", "2025-09-24 13:37:43,978 - INFO - Batch 308/1250\n", "2025-09-24 13:37:44,041 - INFO - Batch 309/1250\n", "2025-09-24 13:37:44,103 - INFO - Batch 310/1250\n", "2025-09-24 13:37:44,165 - INFO - Batch 311/1250\n", "2025-09-24 13:37:44,228 - INFO - Batch 312/1250\n", "2025-09-24 13:37:44,293 - INFO - Batch 313/1250\n", "2025-09-24 13:37:44,351 - INFO - Batch 314/1250\n", "2025-09-24 13:37:44,411 - INFO - Batch 315/1250\n", "2025-09-24 13:37:44,473 - INFO - Batch 316/1250\n", "2025-09-24 13:37:44,537 - INFO - Batch 317/1250\n", "2025-09-24 13:37:44,598 - INFO - Batch 318/1250\n", "2025-09-24 13:37:44,661 - INFO - Batch 319/1250\n", "2025-09-24 13:37:44,731 - INFO - Batch 320/1250\n", "2025-09-24 13:37:44,796 - INFO - Batch 321/1250\n", "2025-09-24 13:37:44,854 - INFO - Batch 322/1250\n", "2025-09-24 13:37:44,919 - INFO - Batch 323/1250\n", "2025-09-24 13:37:44,978 - INFO - Batch 324/1250\n", "2025-09-24 13:37:45,038 - INFO - Batch 325/1250\n", "2025-09-24 13:37:45,099 - INFO - Batch 326/1250\n", "2025-09-24 13:37:45,163 - INFO - Batch 327/1250\n", "2025-09-24 13:37:45,224 - INFO - Batch 328/1250\n", "2025-09-24 13:37:45,286 - INFO - Batch 329/1250\n", "2025-09-24 13:37:45,346 - INFO - Batch 330/1250\n", "2025-09-24 13:37:45,405 - INFO - Batch 331/1250\n", "2025-09-24 13:37:45,464 - INFO - Batch 332/1250\n", "2025-09-24 13:37:45,522 - INFO - Batch 333/1250\n", "2025-09-24 13:37:45,586 - INFO - Batch 334/1250\n", "2025-09-24 13:37:45,646 - INFO - Batch 335/1250\n", "2025-09-24 13:37:45,704 - INFO - Batch 336/1250\n", "2025-09-24 13:37:45,763 - INFO - Batch 337/1250\n", "2025-09-24 13:37:45,827 - INFO - Batch 338/1250\n", "2025-09-24 13:37:45,888 - INFO - Batch 339/1250\n", "2025-09-24 13:37:45,950 - INFO - Batch 340/1250\n", "2025-09-24 13:37:46,015 - INFO - Batch 341/1250\n", "2025-09-24 13:37:46,080 - INFO - Batch 342/1250\n", "2025-09-24 13:37:46,141 - INFO - Batch 343/1250\n", "2025-09-24 13:37:46,204 - INFO - Batch 344/1250\n", "2025-09-24 13:37:46,267 - INFO - Batch 345/1250\n", "2025-09-24 13:37:46,331 - INFO - Batch 346/1250\n", "2025-09-24 13:37:46,394 - INFO - Batch 347/1250\n", "2025-09-24 13:37:46,456 - INFO - Batch 348/1250\n", "2025-09-24 13:37:46,516 - INFO - Batch 349/1250\n", "2025-09-24 13:37:46,608 - INFO - Batch 350/1250\n", "2025-09-24 13:37:46,693 - INFO - Batch 351/1250\n", "2025-09-24 13:37:46,770 - INFO - Batch 352/1250\n", "2025-09-24 13:37:46,841 - INFO - Batch 353/1250\n", "2025-09-24 13:37:46,912 - INFO - Batch 354/1250\n", "2025-09-24 13:37:46,983 - INFO - Batch 355/1250\n", "2025-09-24 13:37:47,057 - INFO - Batch 356/1250\n", "2025-09-24 13:37:47,127 - INFO - Batch 357/1250\n", "2025-09-24 13:37:47,199 - INFO - Batch 358/1250\n", "2025-09-24 13:37:47,275 - INFO - Batch 359/1250\n", "2025-09-24 13:37:47,350 - INFO - Batch 360/1250\n", "2025-09-24 13:37:47,423 - INFO - Batch 361/1250\n", "2025-09-24 13:37:47,494 - INFO - Batch 362/1250\n", "2025-09-24 13:37:47,570 - INFO - Batch 363/1250\n", "2025-09-24 13:37:47,645 - INFO - Batch 364/1250\n", "2025-09-24 13:37:47,723 - INFO - Batch 365/1250\n", "2025-09-24 13:37:47,800 - INFO - Batch 366/1250\n", "2025-09-24 13:37:47,879 - INFO - Batch 367/1250\n", "2025-09-24 13:37:47,950 - INFO - Batch 368/1250\n", "2025-09-24 13:37:48,024 - INFO - Batch 369/1250\n", "2025-09-24 13:37:48,097 - INFO - Batch 370/1250\n", "2025-09-24 13:37:48,172 - INFO - Batch 371/1250\n", "2025-09-24 13:37:48,247 - INFO - Batch 372/1250\n", "2025-09-24 13:37:48,319 - INFO - Batch 373/1250\n", "2025-09-24 13:37:48,392 - INFO - Batch 374/1250\n", "2025-09-24 13:37:48,468 - INFO - Batch 375/1250\n", "2025-09-24 13:37:48,538 - INFO - Batch 376/1250\n", "2025-09-24 13:37:48,612 - INFO - Batch 377/1250\n", "2025-09-24 13:37:48,683 - INFO - Batch 378/1250\n", "2025-09-24 13:37:48,755 - INFO - Batch 379/1250\n", "2025-09-24 13:37:48,829 - INFO - Batch 380/1250\n", "2025-09-24 13:37:48,903 - INFO - Batch 381/1250\n", "2025-09-24 13:37:48,978 - INFO - Batch 382/1250\n", "2025-09-24 13:37:49,050 - INFO - Batch 383/1250\n", "2025-09-24 13:37:49,116 - INFO - Batch 384/1250\n", "2025-09-24 13:37:49,176 - INFO - Batch 385/1250\n", "2025-09-24 13:37:49,240 - INFO - Batch 386/1250\n", "2025-09-24 13:37:49,307 - INFO - Batch 387/1250\n", "2025-09-24 13:37:49,368 - INFO - Batch 388/1250\n", "2025-09-24 13:37:49,430 - INFO - Batch 389/1250\n", "2025-09-24 13:37:49,492 - INFO - Batch 390/1250\n", "2025-09-24 13:37:49,553 - INFO - Batch 391/1250\n", "2025-09-24 13:37:49,616 - INFO - Batch 392/1250\n", "2025-09-24 13:37:49,681 - INFO - Batch 393/1250\n", "2025-09-24 13:37:49,745 - INFO - Batch 394/1250\n", "2025-09-24 13:37:49,803 - INFO - Batch 395/1250\n", "2025-09-24 13:37:49,862 - INFO - Batch 396/1250\n", "2025-09-24 13:37:49,925 - INFO - Batch 397/1250\n", "2025-09-24 13:37:49,999 - INFO - Batch 398/1250\n", "2025-09-24 13:37:50,059 - INFO - Batch 399/1250\n", "2025-09-24 13:37:50,120 - INFO - Batch 400/1250\n", "2025-09-24 13:37:50,181 - INFO - Batch 401/1250\n", "2025-09-24 13:37:50,245 - INFO - Batch 402/1250\n", "2025-09-24 13:37:50,306 - INFO - Batch 403/1250\n", "2025-09-24 13:37:50,370 - INFO - Batch 404/1250\n", "2025-09-24 13:37:50,433 - INFO - Batch 405/1250\n", "2025-09-24 13:37:50,497 - INFO - Batch 406/1250\n", "2025-09-24 13:37:50,561 - INFO - Batch 407/1250\n", "2025-09-24 13:37:50,619 - INFO - Batch 408/1250\n", "2025-09-24 13:37:50,683 - INFO - Batch 409/1250\n", "2025-09-24 13:37:50,745 - INFO - Batch 410/1250\n", "2025-09-24 13:37:50,806 - INFO - Batch 411/1250\n", "2025-09-24 13:37:50,866 - INFO - Batch 412/1250\n", "2025-09-24 13:37:50,927 - INFO - Batch 413/1250\n", "2025-09-24 13:37:50,990 - INFO - Batch 414/1250\n", "2025-09-24 13:37:51,048 - INFO - Batch 415/1250\n", "2025-09-24 13:37:51,110 - INFO - Batch 416/1250\n", "2025-09-24 13:37:51,173 - INFO - Batch 417/1250\n", "2025-09-24 13:37:51,234 - INFO - Batch 418/1250\n", "2025-09-24 13:37:51,296 - INFO - Batch 419/1250\n", "2025-09-24 13:37:51,362 - INFO - Batch 420/1250\n", "2025-09-24 13:37:51,424 - INFO - Batch 421/1250\n", "2025-09-24 13:37:51,483 - INFO - Batch 422/1250\n", "2025-09-24 13:37:51,545 - INFO - Batch 423/1250\n", "2025-09-24 13:37:51,607 - INFO - Batch 424/1250\n", "2025-09-24 13:37:51,670 - INFO - Batch 425/1250\n", "2025-09-24 13:37:51,731 - INFO - Batch 426/1250\n", "2025-09-24 13:37:51,791 - INFO - Batch 427/1250\n", "2025-09-24 13:37:51,851 - INFO - Batch 428/1250\n", "2025-09-24 13:37:51,912 - INFO - Batch 429/1250\n", "2025-09-24 13:37:51,972 - INFO - Batch 430/1250\n", "2025-09-24 13:37:52,032 - INFO - Batch 431/1250\n", "2025-09-24 13:37:52,092 - INFO - Batch 432/1250\n", "2025-09-24 13:37:52,150 - INFO - Batch 433/1250\n", "2025-09-24 13:37:52,211 - INFO - Batch 434/1250\n", "2025-09-24 13:37:52,272 - INFO - Batch 435/1250\n", "2025-09-24 13:37:52,330 - INFO - Batch 436/1250\n", "2025-09-24 13:37:52,390 - INFO - Batch 437/1250\n", "2025-09-24 13:37:52,450 - INFO - Batch 438/1250\n", "2025-09-24 13:37:52,508 - INFO - Batch 439/1250\n", "2025-09-24 13:37:52,573 - INFO - Batch 440/1250\n", "2025-09-24 13:37:52,632 - INFO - Batch 441/1250\n", "2025-09-24 13:37:52,691 - INFO - Batch 442/1250\n", "2025-09-24 13:37:52,753 - INFO - Batch 443/1250\n", "2025-09-24 13:37:52,816 - INFO - Batch 444/1250\n", "2025-09-24 13:37:52,876 - INFO - Batch 445/1250\n", "2025-09-24 13:37:52,939 - INFO - Batch 446/1250\n", "2025-09-24 13:37:52,999 - INFO - Batch 447/1250\n", "2025-09-24 13:37:53,062 - INFO - Batch 448/1250\n", "2025-09-24 13:37:53,125 - INFO - Batch 449/1250\n", "2025-09-24 13:37:53,189 - INFO - Batch 450/1250\n", "2025-09-24 13:37:53,249 - INFO - Batch 451/1250\n", "2025-09-24 13:37:53,313 - INFO - Batch 452/1250\n", "2025-09-24 13:37:53,375 - INFO - Batch 453/1250\n", "2025-09-24 13:37:53,438 - INFO - Batch 454/1250\n", "2025-09-24 13:37:53,500 - INFO - Batch 455/1250\n", "2025-09-24 13:37:53,560 - INFO - Batch 456/1250\n", "2025-09-24 13:37:53,621 - INFO - Batch 457/1250\n", "2025-09-24 13:37:53,687 - INFO - Batch 458/1250\n", "2025-09-24 13:37:53,750 - INFO - Batch 459/1250\n", "2025-09-24 13:37:53,811 - INFO - Batch 460/1250\n", "2025-09-24 13:37:53,872 - INFO - Batch 461/1250\n", "2025-09-24 13:37:53,939 - INFO - Batch 462/1250\n", "2025-09-24 13:37:53,999 - INFO - Batch 463/1250\n", "2025-09-24 13:37:54,057 - INFO - Batch 464/1250\n", "2025-09-24 13:37:54,122 - INFO - Batch 465/1250\n", "2025-09-24 13:37:54,186 - INFO - Batch 466/1250\n", "2025-09-24 13:37:54,245 - INFO - Batch 467/1250\n", "2025-09-24 13:37:54,309 - INFO - Batch 468/1250\n", "2025-09-24 13:37:54,370 - INFO - Batch 469/1250\n", "2025-09-24 13:37:54,433 - INFO - Batch 470/1250\n", "2025-09-24 13:37:54,493 - INFO - Batch 471/1250\n", "2025-09-24 13:37:54,555 - INFO - Batch 472/1250\n", "2025-09-24 13:37:54,620 - INFO - Batch 473/1250\n", "2025-09-24 13:37:54,685 - INFO - Batch 474/1250\n", "2025-09-24 13:37:54,742 - INFO - Batch 475/1250\n", "2025-09-24 13:37:54,811 - INFO - Batch 476/1250\n", "2025-09-24 13:37:54,872 - INFO - Batch 477/1250\n", "2025-09-24 13:37:54,932 - INFO - Batch 478/1250\n", "2025-09-24 13:37:54,992 - INFO - Batch 479/1250\n", "2025-09-24 13:37:55,054 - INFO - Batch 480/1250\n", "2025-09-24 13:37:55,116 - INFO - Batch 481/1250\n", "2025-09-24 13:37:55,178 - INFO - Batch 482/1250\n", "2025-09-24 13:37:55,243 - INFO - Batch 483/1250\n", "2025-09-24 13:37:55,305 - INFO - Batch 484/1250\n", "2025-09-24 13:37:55,369 - INFO - Batch 485/1250\n", "2025-09-24 13:37:55,434 - INFO - Batch 486/1250\n", "2025-09-24 13:37:55,495 - INFO - Batch 487/1250\n", "2025-09-24 13:37:55,559 - INFO - Batch 488/1250\n", "2025-09-24 13:37:55,621 - INFO - Batch 489/1250\n", "2025-09-24 13:37:55,682 - INFO - Batch 490/1250\n", "2025-09-24 13:37:55,746 - INFO - Batch 491/1250\n", "2025-09-24 13:37:55,808 - INFO - Batch 492/1250\n", "2025-09-24 13:37:55,866 - INFO - Batch 493/1250\n", "2025-09-24 13:37:55,931 - INFO - Batch 494/1250\n", "2025-09-24 13:37:55,995 - INFO - Batch 495/1250\n", "2025-09-24 13:37:56,056 - INFO - Batch 496/1250\n", "2025-09-24 13:37:56,115 - INFO - Batch 497/1250\n", "2025-09-24 13:37:56,175 - INFO - Batch 498/1250\n", "2025-09-24 13:37:56,238 - INFO - Batch 499/1250\n", "2025-09-24 13:37:56,303 - INFO - Batch 500/1250\n", "2025-09-24 13:37:56,365 - INFO - Batch 501/1250\n", "2025-09-24 13:37:56,425 - INFO - Batch 502/1250\n", "2025-09-24 13:37:56,488 - INFO - Batch 503/1250\n", "2025-09-24 13:37:56,550 - INFO - Batch 504/1250\n", "2025-09-24 13:37:56,613 - INFO - Batch 505/1250\n", "2025-09-24 13:37:56,677 - INFO - Batch 506/1250\n", "2025-09-24 13:37:56,738 - INFO - Batch 507/1250\n", "2025-09-24 13:37:56,803 - INFO - Batch 508/1250\n", "2025-09-24 13:37:56,864 - INFO - Batch 509/1250\n", "2025-09-24 13:37:56,925 - INFO - Batch 510/1250\n", "2025-09-24 13:37:56,987 - INFO - Batch 511/1250\n", "2025-09-24 13:37:57,047 - INFO - Batch 512/1250\n", "2025-09-24 13:37:57,107 - INFO - Batch 513/1250\n", "2025-09-24 13:37:57,167 - INFO - Batch 514/1250\n", "2025-09-24 13:37:57,227 - INFO - Batch 515/1250\n", "2025-09-24 13:37:57,290 - INFO - Batch 516/1250\n", "2025-09-24 13:37:57,355 - INFO - Batch 517/1250\n", "2025-09-24 13:37:57,418 - INFO - Batch 518/1250\n", "2025-09-24 13:37:57,481 - INFO - Batch 519/1250\n", "2025-09-24 13:37:57,545 - INFO - Batch 520/1250\n", "2025-09-24 13:37:57,613 - INFO - Batch 521/1250\n", "2025-09-24 13:37:57,680 - INFO - Batch 522/1250\n", "2025-09-24 13:37:57,740 - INFO - Batch 523/1250\n", "2025-09-24 13:37:57,798 - INFO - Batch 524/1250\n", "2025-09-24 13:37:57,860 - INFO - Batch 525/1250\n", "2025-09-24 13:37:57,929 - INFO - Batch 526/1250\n", "2025-09-24 13:37:57,989 - INFO - Batch 527/1250\n", "2025-09-24 13:37:58,053 - INFO - Batch 528/1250\n", "2025-09-24 13:37:58,118 - INFO - Batch 529/1250\n", "2025-09-24 13:37:58,180 - INFO - Batch 530/1250\n", "2025-09-24 13:37:58,241 - INFO - Batch 531/1250\n", "2025-09-24 13:37:58,302 - INFO - Batch 532/1250\n", "2025-09-24 13:37:58,361 - INFO - Batch 533/1250\n", "2025-09-24 13:37:58,424 - INFO - Batch 534/1250\n", "2025-09-24 13:37:58,490 - INFO - Batch 535/1250\n", "2025-09-24 13:37:58,556 - INFO - Batch 536/1250\n", "2025-09-24 13:37:58,616 - INFO - Batch 537/1250\n", "2025-09-24 13:37:58,678 - INFO - Batch 538/1250\n", "2025-09-24 13:37:58,739 - INFO - Batch 539/1250\n", "2025-09-24 13:37:58,808 - INFO - Batch 540/1250\n", "2025-09-24 13:37:58,872 - INFO - Batch 541/1250\n", "2025-09-24 13:37:58,936 - INFO - Batch 542/1250\n", "2025-09-24 13:37:58,998 - INFO - Batch 543/1250\n", "2025-09-24 13:37:59,065 - INFO - Batch 544/1250\n", "2025-09-24 13:37:59,131 - INFO - Batch 545/1250\n", "2025-09-24 13:37:59,192 - INFO - Batch 546/1250\n", "2025-09-24 13:37:59,254 - INFO - Batch 547/1250\n", "2025-09-24 13:37:59,319 - INFO - Batch 548/1250\n", "2025-09-24 13:37:59,386 - INFO - Batch 549/1250\n", "2025-09-24 13:37:59,452 - INFO - Batch 550/1250\n", "2025-09-24 13:37:59,515 - INFO - Batch 551/1250\n", "2025-09-24 13:37:59,578 - INFO - Batch 552/1250\n", "2025-09-24 13:37:59,641 - INFO - Batch 553/1250\n", "2025-09-24 13:37:59,704 - INFO - Batch 554/1250\n", "2025-09-24 13:37:59,767 - INFO - Batch 555/1250\n", "2025-09-24 13:37:59,829 - INFO - Batch 556/1250\n", "2025-09-24 13:37:59,891 - INFO - Batch 557/1250\n", "2025-09-24 13:37:59,957 - INFO - Batch 558/1250\n", "2025-09-24 13:38:00,021 - INFO - Batch 559/1250\n", "2025-09-24 13:38:00,084 - INFO - Batch 560/1250\n", "2025-09-24 13:38:00,153 - INFO - Batch 561/1250\n", "2025-09-24 13:38:00,250 - INFO - Batch 562/1250\n", "2025-09-24 13:38:00,321 - INFO - Batch 563/1250\n", "2025-09-24 13:38:00,384 - INFO - Batch 564/1250\n", "2025-09-24 13:38:00,445 - INFO - Batch 565/1250\n", "2025-09-24 13:38:00,505 - INFO - Batch 566/1250\n", "2025-09-24 13:38:00,570 - INFO - Batch 567/1250\n", "2025-09-24 13:38:00,633 - INFO - Batch 568/1250\n", "2025-09-24 13:38:00,693 - INFO - Batch 569/1250\n", "2025-09-24 13:38:00,754 - INFO - Batch 570/1250\n", "2025-09-24 13:38:00,814 - INFO - Batch 571/1250\n", "2025-09-24 13:38:00,878 - INFO - Batch 572/1250\n", "2025-09-24 13:38:00,940 - INFO - Batch 573/1250\n", "2025-09-24 13:38:01,002 - INFO - Batch 574/1250\n", "2025-09-24 13:38:01,067 - INFO - Batch 575/1250\n", "2025-09-24 13:38:01,133 - INFO - Batch 576/1250\n", "2025-09-24 13:38:01,190 - INFO - Batch 577/1250\n", "2025-09-24 13:38:01,250 - INFO - Batch 578/1250\n", "2025-09-24 13:38:01,310 - INFO - Batch 579/1250\n", "2025-09-24 13:38:01,372 - INFO - Batch 580/1250\n", "2025-09-24 13:38:01,432 - INFO - Batch 581/1250\n", "2025-09-24 13:38:01,491 - INFO - Batch 582/1250\n", "2025-09-24 13:38:01,570 - INFO - Batch 583/1250\n", "2025-09-24 13:38:01,632 - INFO - Batch 584/1250\n", "2025-09-24 13:38:01,692 - INFO - Batch 585/1250\n", "2025-09-24 13:38:01,756 - INFO - Batch 586/1250\n", "2025-09-24 13:38:01,815 - INFO - Batch 587/1250\n", "2025-09-24 13:38:01,876 - INFO - Batch 588/1250\n", "2025-09-24 13:38:01,940 - INFO - Batch 589/1250\n", "2025-09-24 13:38:02,002 - INFO - Batch 590/1250\n", "2025-09-24 13:38:02,066 - INFO - Batch 591/1250\n", "2025-09-24 13:38:02,125 - INFO - Batch 592/1250\n", "2025-09-24 13:38:02,186 - INFO - Batch 593/1250\n", "2025-09-24 13:38:02,251 - INFO - Batch 594/1250\n", "2025-09-24 13:38:02,313 - INFO - Batch 595/1250\n", "2025-09-24 13:38:02,373 - INFO - Batch 596/1250\n", "2025-09-24 13:38:02,434 - INFO - Batch 597/1250\n", "2025-09-24 13:38:02,495 - INFO - Batch 598/1250\n", "2025-09-24 13:38:02,558 - INFO - Batch 599/1250\n", "2025-09-24 13:38:02,622 - INFO - Batch 600/1250\n", "2025-09-24 13:38:02,678 - INFO - Batch 601/1250\n", "2025-09-24 13:38:02,742 - INFO - Batch 602/1250\n", "2025-09-24 13:38:02,807 - INFO - Batch 603/1250\n", "2025-09-24 13:38:02,870 - INFO - Batch 604/1250\n", "2025-09-24 13:38:02,930 - INFO - Batch 605/1250\n", "2025-09-24 13:38:02,992 - INFO - Batch 606/1250\n", "2025-09-24 13:38:03,051 - INFO - Batch 607/1250\n", "2025-09-24 13:38:03,115 - INFO - Batch 608/1250\n", "2025-09-24 13:38:03,176 - INFO - Batch 609/1250\n", "2025-09-24 13:38:03,234 - INFO - Batch 610/1250\n", "2025-09-24 13:38:03,298 - INFO - Batch 611/1250\n", "2025-09-24 13:38:03,361 - INFO - Batch 612/1250\n", "2025-09-24 13:38:03,423 - INFO - Batch 613/1250\n", "2025-09-24 13:38:03,486 - INFO - Batch 614/1250\n", "2025-09-24 13:38:03,547 - INFO - Batch 615/1250\n", "2025-09-24 13:38:03,609 - INFO - Batch 616/1250\n", "2025-09-24 13:38:03,670 - INFO - Batch 617/1250\n", "2025-09-24 13:38:03,732 - INFO - Batch 618/1250\n", "2025-09-24 13:38:03,795 - INFO - Batch 619/1250\n", "2025-09-24 13:38:03,854 - INFO - Batch 620/1250\n", "2025-09-24 13:38:03,916 - INFO - Batch 621/1250\n", "2025-09-24 13:38:03,977 - INFO - Batch 622/1250\n", "2025-09-24 13:38:04,040 - INFO - Batch 623/1250\n", "2025-09-24 13:38:04,102 - INFO - Batch 624/1250\n", "2025-09-24 13:38:04,161 - INFO - Batch 625/1250\n", "2025-09-24 13:38:04,218 - INFO - Batch 626/1250\n", "2025-09-24 13:38:04,280 - INFO - Batch 627/1250\n", "2025-09-24 13:38:04,340 - INFO - Batch 628/1250\n", "2025-09-24 13:38:04,405 - INFO - Batch 629/1250\n", "2025-09-24 13:38:04,469 - INFO - Batch 630/1250\n", "2025-09-24 13:38:04,531 - INFO - Batch 631/1250\n", "2025-09-24 13:38:04,595 - INFO - Batch 632/1250\n", "2025-09-24 13:38:04,656 - INFO - Batch 633/1250\n", "2025-09-24 13:38:04,718 - INFO - Batch 634/1250\n", "2025-09-24 13:38:04,783 - INFO - Batch 635/1250\n", "2025-09-24 13:38:04,845 - INFO - Batch 636/1250\n", "2025-09-24 13:38:04,910 - INFO - Batch 637/1250\n", "2025-09-24 13:38:04,971 - INFO - Batch 638/1250\n", "2025-09-24 13:38:05,033 - INFO - Batch 639/1250\n", "2025-09-24 13:38:05,100 - INFO - Batch 640/1250\n", "2025-09-24 13:38:05,164 - INFO - Batch 641/1250\n", "2025-09-24 13:38:05,225 - INFO - Batch 642/1250\n", "2025-09-24 13:38:05,290 - INFO - Batch 643/1250\n", "2025-09-24 13:38:05,356 - INFO - Batch 644/1250\n", "2025-09-24 13:38:05,420 - INFO - Batch 645/1250\n", "2025-09-24 13:38:05,481 - INFO - Batch 646/1250\n", "2025-09-24 13:38:05,547 - INFO - Batch 647/1250\n", "2025-09-24 13:38:05,612 - INFO - Batch 648/1250\n", "2025-09-24 13:38:05,678 - INFO - Batch 649/1250\n", "2025-09-24 13:38:05,738 - INFO - Batch 650/1250\n", "2025-09-24 13:38:05,797 - INFO - Batch 651/1250\n", "2025-09-24 13:38:05,857 - INFO - Batch 652/1250\n", "2025-09-24 13:38:05,925 - INFO - Batch 653/1250\n", "2025-09-24 13:38:05,990 - INFO - Batch 654/1250\n", "2025-09-24 13:38:06,054 - INFO - Batch 655/1250\n", "2025-09-24 13:38:06,116 - INFO - Batch 656/1250\n", "2025-09-24 13:38:06,181 - INFO - Batch 657/1250\n", "2025-09-24 13:38:06,241 - INFO - Batch 658/1250\n", "2025-09-24 13:38:06,300 - INFO - Batch 659/1250\n", "2025-09-24 13:38:06,366 - INFO - Batch 660/1250\n", "2025-09-24 13:38:06,431 - INFO - Batch 661/1250\n", "2025-09-24 13:38:06,494 - INFO - Batch 662/1250\n", "2025-09-24 13:38:06,563 - INFO - Batch 663/1250\n", "2025-09-24 13:38:06,626 - INFO - Batch 664/1250\n", "2025-09-24 13:38:06,689 - INFO - Batch 665/1250\n", "2025-09-24 13:38:06,749 - INFO - Batch 666/1250\n", "2025-09-24 13:38:06,812 - INFO - Batch 667/1250\n", "2025-09-24 13:38:06,876 - INFO - Batch 668/1250\n", "2025-09-24 13:38:06,936 - INFO - Batch 669/1250\n", "2025-09-24 13:38:06,999 - INFO - Batch 670/1250\n", "2025-09-24 13:38:07,062 - INFO - Batch 671/1250\n", "2025-09-24 13:38:07,127 - INFO - Batch 672/1250\n", "2025-09-24 13:38:07,188 - INFO - Batch 673/1250\n", "2025-09-24 13:38:07,249 - INFO - Batch 674/1250\n", "2025-09-24 13:38:07,312 - INFO - Batch 675/1250\n", "2025-09-24 13:38:07,377 - INFO - Batch 676/1250\n", "2025-09-24 13:38:07,440 - INFO - Batch 677/1250\n", "2025-09-24 13:38:07,503 - INFO - Batch 678/1250\n", "2025-09-24 13:38:07,563 - INFO - Batch 679/1250\n", "2025-09-24 13:38:07,634 - INFO - Batch 680/1250\n", "2025-09-24 13:38:07,701 - INFO - Batch 681/1250\n", "2025-09-24 13:38:07,764 - INFO - Batch 682/1250\n", "2025-09-24 13:38:07,828 - INFO - Batch 683/1250\n", "2025-09-24 13:38:07,896 - INFO - Batch 684/1250\n", "2025-09-24 13:38:07,960 - INFO - Batch 685/1250\n", "2025-09-24 13:38:08,020 - INFO - Batch 686/1250\n", "2025-09-24 13:38:08,086 - INFO - Batch 687/1250\n", "2025-09-24 13:38:08,143 - INFO - Batch 688/1250\n", "2025-09-24 13:38:08,205 - INFO - Batch 689/1250\n", "2025-09-24 13:38:08,270 - INFO - Batch 690/1250\n", "2025-09-24 13:38:08,332 - INFO - Batch 691/1250\n", "2025-09-24 13:38:08,396 - INFO - Batch 692/1250\n", "2025-09-24 13:38:08,491 - INFO - Batch 693/1250\n", "2025-09-24 13:38:08,553 - INFO - Batch 694/1250\n", "2025-09-24 13:38:08,620 - INFO - Batch 695/1250\n", "2025-09-24 13:38:08,680 - INFO - Batch 696/1250\n", "2025-09-24 13:38:08,739 - INFO - Batch 697/1250\n", "2025-09-24 13:38:08,804 - INFO - Batch 698/1250\n", "2025-09-24 13:38:08,866 - INFO - Batch 699/1250\n", "2025-09-24 13:38:08,929 - INFO - Batch 700/1250\n", "2025-09-24 13:38:08,999 - INFO - Batch 701/1250\n", "2025-09-24 13:38:09,060 - INFO - Batch 702/1250\n", "2025-09-24 13:38:09,123 - INFO - Batch 703/1250\n", "2025-09-24 13:38:09,193 - INFO - Batch 704/1250\n", "2025-09-24 13:38:09,255 - INFO - Batch 705/1250\n", "2025-09-24 13:38:09,313 - INFO - Batch 706/1250\n", "2025-09-24 13:38:09,378 - INFO - Batch 707/1250\n", "2025-09-24 13:38:09,451 - INFO - Batch 708/1250\n", "2025-09-24 13:38:09,514 - INFO - Batch 709/1250\n", "2025-09-24 13:38:09,577 - INFO - Batch 710/1250\n", "2025-09-24 13:38:09,639 - INFO - Batch 711/1250\n", "2025-09-24 13:38:09,701 - INFO - Batch 712/1250\n", "2025-09-24 13:38:09,764 - INFO - Batch 713/1250\n", "2025-09-24 13:38:09,826 - INFO - Batch 714/1250\n", "2025-09-24 13:38:09,890 - INFO - Batch 715/1250\n", "2025-09-24 13:38:09,952 - INFO - Batch 716/1250\n", "2025-09-24 13:38:10,018 - INFO - Batch 717/1250\n", "2025-09-24 13:38:10,078 - INFO - Batch 718/1250\n", "2025-09-24 13:38:10,137 - INFO - Batch 719/1250\n", "2025-09-24 13:38:10,198 - INFO - Batch 720/1250\n", "2025-09-24 13:38:10,259 - INFO - Batch 721/1250\n", "2025-09-24 13:38:10,325 - INFO - Batch 722/1250\n", "2025-09-24 13:38:10,390 - INFO - Batch 723/1250\n", "2025-09-24 13:38:10,448 - INFO - Batch 724/1250\n", "2025-09-24 13:38:10,513 - INFO - Batch 725/1250\n", "2025-09-24 13:38:10,578 - INFO - Batch 726/1250\n", "2025-09-24 13:38:10,636 - INFO - Batch 727/1250\n", "2025-09-24 13:38:10,704 - INFO - Batch 728/1250\n", "2025-09-24 13:38:10,765 - INFO - Batch 729/1250\n", "2025-09-24 13:38:10,829 - INFO - Batch 730/1250\n", "2025-09-24 13:38:10,892 - INFO - Batch 731/1250\n", "2025-09-24 13:38:10,955 - INFO - Batch 732/1250\n", "2025-09-24 13:38:11,017 - INFO - Batch 733/1250\n", "2025-09-24 13:38:11,078 - INFO - Batch 734/1250\n", "2025-09-24 13:38:11,142 - INFO - Batch 735/1250\n", "2025-09-24 13:38:11,205 - INFO - Batch 736/1250\n", "2025-09-24 13:38:11,268 - INFO - Batch 737/1250\n", "2025-09-24 13:38:11,326 - INFO - Batch 738/1250\n", "2025-09-24 13:38:11,390 - INFO - Batch 739/1250\n", "2025-09-24 13:38:11,453 - INFO - Batch 740/1250\n", "2025-09-24 13:38:11,517 - INFO - Batch 741/1250\n", "2025-09-24 13:38:11,583 - INFO - Batch 742/1250\n", "2025-09-24 13:38:11,650 - INFO - Batch 743/1250\n", "2025-09-24 13:38:11,717 - INFO - Batch 744/1250\n", "2025-09-24 13:38:11,787 - INFO - Batch 745/1250\n", "2025-09-24 13:38:11,854 - INFO - Batch 746/1250\n", "2025-09-24 13:38:11,913 - INFO - Batch 747/1250\n", "2025-09-24 13:38:11,975 - INFO - Batch 748/1250\n", "2025-09-24 13:38:12,041 - INFO - Batch 749/1250\n", "2025-09-24 13:38:12,106 - INFO - Batch 750/1250\n", "2025-09-24 13:38:12,168 - INFO - Batch 751/1250\n", "2025-09-24 13:38:12,230 - INFO - Batch 752/1250\n", "2025-09-24 13:38:12,295 - INFO - Batch 753/1250\n", "2025-09-24 13:38:12,359 - INFO - Batch 754/1250\n", "2025-09-24 13:38:12,423 - INFO - Batch 755/1250\n", "2025-09-24 13:38:12,492 - INFO - Batch 756/1250\n", "2025-09-24 13:38:12,557 - INFO - Batch 757/1250\n", "2025-09-24 13:38:12,623 - INFO - Batch 758/1250\n", "2025-09-24 13:38:12,687 - INFO - Batch 759/1250\n", "2025-09-24 13:38:12,753 - INFO - Batch 760/1250\n", "2025-09-24 13:38:12,812 - INFO - Batch 761/1250\n", "2025-09-24 13:38:12,870 - INFO - Batch 762/1250\n", "2025-09-24 13:38:12,936 - INFO - Batch 763/1250\n", "2025-09-24 13:38:13,005 - INFO - Batch 764/1250\n", "2025-09-24 13:38:13,072 - INFO - Batch 765/1250\n", "2025-09-24 13:38:13,137 - INFO - Batch 766/1250\n", "2025-09-24 13:38:13,201 - INFO - Batch 767/1250\n", "2025-09-24 13:38:13,262 - INFO - Batch 768/1250\n", "2025-09-24 13:38:13,327 - INFO - Batch 769/1250\n", "2025-09-24 13:38:13,392 - INFO - Batch 770/1250\n", "2025-09-24 13:38:13,449 - INFO - Batch 771/1250\n", "2025-09-24 13:38:13,515 - INFO - Batch 772/1250\n", "2025-09-24 13:38:13,583 - INFO - Batch 773/1250\n", "2025-09-24 13:38:13,648 - INFO - Batch 774/1250\n", "2025-09-24 13:38:13,708 - INFO - Batch 775/1250\n", "2025-09-24 13:38:13,770 - INFO - Batch 776/1250\n", "2025-09-24 13:38:13,834 - INFO - Batch 777/1250\n", "2025-09-24 13:38:13,899 - INFO - Batch 778/1250\n", "2025-09-24 13:38:13,962 - INFO - Batch 779/1250\n", "2025-09-24 13:38:14,026 - INFO - Batch 780/1250\n", "2025-09-24 13:38:14,088 - INFO - Batch 781/1250\n", "2025-09-24 13:38:14,149 - INFO - Batch 782/1250\n", "2025-09-24 13:38:14,214 - INFO - Batch 783/1250\n", "2025-09-24 13:38:14,277 - INFO - Batch 784/1250\n", "2025-09-24 13:38:14,340 - INFO - Batch 785/1250\n", "2025-09-24 13:38:14,401 - INFO - Batch 786/1250\n", "2025-09-24 13:38:14,461 - INFO - Batch 787/1250\n", "2025-09-24 13:38:14,526 - INFO - Batch 788/1250\n", "2025-09-24 13:38:14,595 - INFO - Batch 789/1250\n", "2025-09-24 13:38:14,661 - INFO - Batch 790/1250\n", "2025-09-24 13:38:14,723 - INFO - Batch 791/1250\n", "2025-09-24 13:38:14,786 - INFO - Batch 792/1250\n", "2025-09-24 13:38:14,849 - INFO - Batch 793/1250\n", "2025-09-24 13:38:14,911 - INFO - Batch 794/1250\n", "2025-09-24 13:38:14,970 - INFO - Batch 795/1250\n", "2025-09-24 13:38:15,033 - INFO - Batch 796/1250\n", "2025-09-24 13:38:15,094 - INFO - Batch 797/1250\n", "2025-09-24 13:38:15,159 - INFO - Batch 798/1250\n", "2025-09-24 13:38:15,222 - INFO - Batch 799/1250\n", "2025-09-24 13:38:15,287 - INFO - Batch 800/1250\n", "2025-09-24 13:38:15,353 - INFO - Batch 801/1250\n", "2025-09-24 13:38:15,421 - INFO - Batch 802/1250\n", "2025-09-24 13:38:15,486 - INFO - Batch 803/1250\n", "2025-09-24 13:38:15,550 - INFO - Batch 804/1250\n", "2025-09-24 13:38:15,612 - INFO - Batch 805/1250\n", "2025-09-24 13:38:15,675 - INFO - Batch 806/1250\n", "2025-09-24 13:38:15,739 - INFO - Batch 807/1250\n", "2025-09-24 13:38:15,800 - INFO - Batch 808/1250\n", "2025-09-24 13:38:15,862 - INFO - Batch 809/1250\n", "2025-09-24 13:38:15,920 - INFO - Batch 810/1250\n", "2025-09-24 13:38:15,980 - INFO - Batch 811/1250\n", "2025-09-24 13:38:16,042 - INFO - Batch 812/1250\n", "2025-09-24 13:38:16,101 - INFO - Batch 813/1250\n", "2025-09-24 13:38:16,166 - INFO - Batch 814/1250\n", "2025-09-24 13:38:16,233 - INFO - Batch 815/1250\n", "2025-09-24 13:38:16,295 - INFO - Batch 816/1250\n", "2025-09-24 13:38:16,362 - INFO - Batch 817/1250\n", "2025-09-24 13:38:16,428 - INFO - Batch 818/1250\n", "2025-09-24 13:38:16,490 - INFO - Batch 819/1250\n", "2025-09-24 13:38:16,557 - INFO - Batch 820/1250\n", "2025-09-24 13:38:16,627 - INFO - Batch 821/1250\n", "2025-09-24 13:38:16,696 - INFO - Batch 822/1250\n", "2025-09-24 13:38:16,759 - INFO - Batch 823/1250\n", "2025-09-24 13:38:16,820 - INFO - Batch 824/1250\n", "2025-09-24 13:38:16,883 - INFO - Batch 825/1250\n", "2025-09-24 13:38:16,948 - INFO - Batch 826/1250\n", "2025-09-24 13:38:17,017 - INFO - Batch 827/1250\n", "2025-09-24 13:38:17,081 - INFO - Batch 828/1250\n", "2025-09-24 13:38:17,142 - INFO - Batch 829/1250\n", "2025-09-24 13:38:17,202 - INFO - Batch 830/1250\n", "2025-09-24 13:38:17,268 - INFO - Batch 831/1250\n", "2025-09-24 13:38:17,328 - INFO - Batch 832/1250\n", "2025-09-24 13:38:17,391 - INFO - Batch 833/1250\n", "2025-09-24 13:38:17,453 - INFO - Batch 834/1250\n", "2025-09-24 13:38:17,514 - INFO - Batch 835/1250\n", "2025-09-24 13:38:17,576 - INFO - Batch 836/1250\n", "2025-09-24 13:38:17,640 - INFO - Batch 837/1250\n", "2025-09-24 13:38:17,702 - INFO - Batch 838/1250\n", "2025-09-24 13:38:17,764 - INFO - Batch 839/1250\n", "2025-09-24 13:38:17,826 - INFO - Batch 840/1250\n", "2025-09-24 13:38:17,890 - INFO - Batch 841/1250\n", "2025-09-24 13:38:17,955 - INFO - Batch 842/1250\n", "2025-09-24 13:38:18,019 - INFO - Batch 843/1250\n", "2025-09-24 13:38:18,085 - INFO - Batch 844/1250\n", "2025-09-24 13:38:18,145 - INFO - Batch 845/1250\n", "2025-09-24 13:38:18,207 - INFO - Batch 846/1250\n", "2025-09-24 13:38:18,269 - INFO - Batch 847/1250\n", "2025-09-24 13:38:18,338 - INFO - Batch 848/1250\n", "2025-09-24 13:38:18,400 - INFO - Batch 849/1250\n", "2025-09-24 13:38:18,470 - INFO - Batch 850/1250\n", "2025-09-24 13:38:18,530 - INFO - Batch 851/1250\n", "2025-09-24 13:38:18,596 - INFO - Batch 852/1250\n", "2025-09-24 13:38:18,659 - INFO - Batch 853/1250\n", "2025-09-24 13:38:18,722 - INFO - Batch 854/1250\n", "2025-09-24 13:38:18,788 - INFO - Batch 855/1250\n", "2025-09-24 13:38:18,853 - INFO - Batch 856/1250\n", "2025-09-24 13:38:18,918 - INFO - Batch 857/1250\n", "2025-09-24 13:38:18,983 - INFO - Batch 858/1250\n", "2025-09-24 13:38:19,044 - INFO - Batch 859/1250\n", "2025-09-24 13:38:19,108 - INFO - Batch 860/1250\n", "2025-09-24 13:38:19,172 - INFO - Batch 861/1250\n", "2025-09-24 13:38:19,238 - INFO - Batch 862/1250\n", "2025-09-24 13:38:19,300 - INFO - Batch 863/1250\n", "2025-09-24 13:38:19,363 - INFO - Batch 864/1250\n", "2025-09-24 13:38:19,425 - INFO - Batch 865/1250\n", "2025-09-24 13:38:19,491 - INFO - Batch 866/1250\n", "2025-09-24 13:38:19,555 - INFO - Batch 867/1250\n", "2025-09-24 13:38:19,623 - INFO - Batch 868/1250\n", "2025-09-24 13:38:19,686 - INFO - Batch 869/1250\n", "2025-09-24 13:38:19,752 - INFO - Batch 870/1250\n", "2025-09-24 13:38:19,816 - INFO - Batch 871/1250\n", "2025-09-24 13:38:19,881 - INFO - Batch 872/1250\n", "2025-09-24 13:38:19,949 - INFO - Batch 873/1250\n", "2025-09-24 13:38:20,015 - INFO - Batch 874/1250\n", "2025-09-24 13:38:20,080 - INFO - Batch 875/1250\n", "2025-09-24 13:38:20,143 - INFO - Batch 876/1250\n", "2025-09-24 13:38:20,198 - INFO - Batch 877/1250\n", "2025-09-24 13:38:20,266 - INFO - Batch 878/1250\n", "2025-09-24 13:38:20,332 - INFO - Batch 879/1250\n", "2025-09-24 13:38:20,394 - INFO - Batch 880/1250\n", "2025-09-24 13:38:20,461 - INFO - Batch 881/1250\n", "2025-09-24 13:38:20,524 - INFO - Batch 882/1250\n", "2025-09-24 13:38:20,586 - INFO - Batch 883/1250\n", "2025-09-24 13:38:20,648 - INFO - Batch 884/1250\n", "2025-09-24 13:38:20,715 - INFO - Batch 885/1250\n", "2025-09-24 13:38:20,779 - INFO - Batch 886/1250\n", "2025-09-24 13:38:20,843 - INFO - Batch 887/1250\n", "2025-09-24 13:38:20,908 - INFO - Batch 888/1250\n", "2025-09-24 13:38:20,972 - INFO - Batch 889/1250\n", "2025-09-24 13:38:21,037 - INFO - Batch 890/1250\n", "2025-09-24 13:38:21,099 - INFO - Batch 891/1250\n", "2025-09-24 13:38:21,163 - INFO - Batch 892/1250\n", "2025-09-24 13:38:21,225 - INFO - Batch 893/1250\n", "2025-09-24 13:38:21,292 - INFO - Batch 894/1250\n", "2025-09-24 13:38:21,353 - INFO - Batch 895/1250\n", "2025-09-24 13:38:21,417 - INFO - Batch 896/1250\n", "2025-09-24 13:38:21,486 - INFO - Batch 897/1250\n", "2025-09-24 13:38:21,550 - INFO - Batch 898/1250\n", "2025-09-24 13:38:21,610 - INFO - Batch 899/1250\n", "2025-09-24 13:38:21,673 - INFO - Batch 900/1250\n", "2025-09-24 13:38:21,740 - INFO - Batch 901/1250\n", "2025-09-24 13:38:21,800 - INFO - Batch 902/1250\n", "2025-09-24 13:38:21,862 - INFO - Batch 903/1250\n", "2025-09-24 13:38:21,928 - INFO - Batch 904/1250\n", "2025-09-24 13:38:21,992 - INFO - Batch 905/1250\n", "2025-09-24 13:38:22,054 - INFO - Batch 906/1250\n", "2025-09-24 13:38:22,112 - INFO - Batch 907/1250\n", "2025-09-24 13:38:22,182 - INFO - Batch 908/1250\n", "2025-09-24 13:38:22,248 - INFO - Batch 909/1250\n", "2025-09-24 13:38:22,314 - INFO - Batch 910/1250\n", "2025-09-24 13:38:22,377 - INFO - Batch 911/1250\n", "2025-09-24 13:38:22,437 - INFO - Batch 912/1250\n", "2025-09-24 13:38:22,501 - INFO - Batch 913/1250\n", "2025-09-24 13:38:22,566 - INFO - Batch 914/1250\n", "2025-09-24 13:38:22,627 - INFO - Batch 915/1250\n", "2025-09-24 13:38:22,687 - INFO - Batch 916/1250\n", "2025-09-24 13:38:22,751 - INFO - Batch 917/1250\n", "2025-09-24 13:38:22,813 - INFO - Batch 918/1250\n", "2025-09-24 13:38:22,875 - INFO - Batch 919/1250\n", "2025-09-24 13:38:22,942 - INFO - Batch 920/1250\n", "2025-09-24 13:38:23,004 - INFO - Batch 921/1250\n", "2025-09-24 13:38:23,069 - INFO - Batch 922/1250\n", "2025-09-24 13:38:23,134 - INFO - Batch 923/1250\n", "2025-09-24 13:38:23,198 - INFO - Batch 924/1250\n", "2025-09-24 13:38:23,269 - INFO - Batch 925/1250\n", "2025-09-24 13:38:23,338 - INFO - Batch 926/1250\n", "2025-09-24 13:38:23,402 - INFO - Batch 927/1250\n", "2025-09-24 13:38:23,467 - INFO - Batch 928/1250\n", "2025-09-24 13:38:23,529 - INFO - Batch 929/1250\n", "2025-09-24 13:38:23,598 - INFO - Batch 930/1250\n", "2025-09-24 13:38:23,667 - INFO - Batch 931/1250\n", "2025-09-24 13:38:23,739 - INFO - Batch 932/1250\n", "2025-09-24 13:38:23,819 - INFO - Batch 933/1250\n", "2025-09-24 13:38:23,883 - INFO - Batch 934/1250\n", "2025-09-24 13:38:23,950 - INFO - Batch 935/1250\n", "2025-09-24 13:38:24,010 - INFO - Batch 936/1250\n", "2025-09-24 13:38:24,073 - INFO - Batch 937/1250\n", "2025-09-24 13:38:24,137 - INFO - Batch 938/1250\n", "2025-09-24 13:38:24,206 - INFO - Batch 939/1250\n", "2025-09-24 13:38:24,271 - INFO - Batch 940/1250\n", "2025-09-24 13:38:24,335 - INFO - Batch 941/1250\n", "2025-09-24 13:38:24,396 - INFO - Batch 942/1250\n", "2025-09-24 13:38:24,460 - INFO - Batch 943/1250\n", "2025-09-24 13:38:24,519 - INFO - Batch 944/1250\n", "2025-09-24 13:38:24,582 - INFO - Batch 945/1250\n", "2025-09-24 13:38:24,647 - INFO - Batch 946/1250\n", "2025-09-24 13:38:24,709 - INFO - Batch 947/1250\n", "2025-09-24 13:38:24,772 - INFO - Batch 948/1250\n", "2025-09-24 13:38:24,835 - INFO - Batch 949/1250\n", "2025-09-24 13:38:24,901 - INFO - Batch 950/1250\n", "2025-09-24 13:38:24,966 - INFO - Batch 951/1250\n", "2025-09-24 13:38:25,027 - INFO - Batch 952/1250\n", "2025-09-24 13:38:25,091 - INFO - Batch 953/1250\n", "2025-09-24 13:38:25,153 - INFO - Batch 954/1250\n", "2025-09-24 13:38:25,216 - INFO - Batch 955/1250\n", "2025-09-24 13:38:25,277 - INFO - Batch 956/1250\n", "2025-09-24 13:38:25,338 - INFO - Batch 957/1250\n", "2025-09-24 13:38:25,402 - INFO - Batch 958/1250\n", "2025-09-24 13:38:25,466 - INFO - Batch 959/1250\n", "2025-09-24 13:38:25,530 - INFO - Batch 960/1250\n", "2025-09-24 13:38:25,593 - INFO - Batch 961/1250\n", "2025-09-24 13:38:25,656 - INFO - Batch 962/1250\n", "2025-09-24 13:38:25,725 - INFO - Batch 963/1250\n", "2025-09-24 13:38:25,787 - INFO - Batch 964/1250\n", "2025-09-24 13:38:25,855 - INFO - Batch 965/1250\n", "2025-09-24 13:38:25,919 - INFO - Batch 966/1250\n", "2025-09-24 13:38:25,982 - INFO - Batch 967/1250\n", "2025-09-24 13:38:26,040 - INFO - Batch 968/1250\n", "2025-09-24 13:38:26,107 - INFO - Batch 969/1250\n", "2025-09-24 13:38:26,176 - INFO - Batch 970/1250\n", "2025-09-24 13:38:26,242 - INFO - Batch 971/1250\n", "2025-09-24 13:38:26,303 - INFO - Batch 972/1250\n", "2025-09-24 13:38:26,370 - INFO - Batch 973/1250\n", "2025-09-24 13:38:26,429 - INFO - Batch 974/1250\n", "2025-09-24 13:38:26,494 - INFO - Batch 975/1250\n", "2025-09-24 13:38:26,556 - INFO - Batch 976/1250\n", "2025-09-24 13:38:26,619 - INFO - Batch 977/1250\n", "2025-09-24 13:38:26,687 - INFO - Batch 978/1250\n", "2025-09-24 13:38:26,746 - INFO - Batch 979/1250\n", "2025-09-24 13:38:26,809 - INFO - Batch 980/1250\n", "2025-09-24 13:38:26,869 - INFO - Batch 981/1250\n", "2025-09-24 13:38:26,930 - INFO - Batch 982/1250\n", "2025-09-24 13:38:26,993 - INFO - Batch 983/1250\n", "2025-09-24 13:38:27,059 - INFO - Batch 984/1250\n", "2025-09-24 13:38:27,123 - INFO - Batch 985/1250\n", "2025-09-24 13:38:27,188 - INFO - Batch 986/1250\n", "2025-09-24 13:38:27,252 - INFO - Batch 987/1250\n", "2025-09-24 13:38:27,315 - INFO - Batch 988/1250\n", "2025-09-24 13:38:27,380 - INFO - Batch 989/1250\n", "2025-09-24 13:38:27,442 - INFO - Batch 990/1250\n", "2025-09-24 13:38:27,504 - INFO - Batch 991/1250\n", "2025-09-24 13:38:27,569 - INFO - Batch 992/1250\n", "2025-09-24 13:38:27,632 - INFO - Batch 993/1250\n", "2025-09-24 13:38:27,702 - INFO - Batch 994/1250\n", "2025-09-24 13:38:27,764 - INFO - Batch 995/1250\n", "2025-09-24 13:38:27,829 - INFO - Batch 996/1250\n", "2025-09-24 13:38:27,895 - INFO - Batch 997/1250\n", "2025-09-24 13:38:27,961 - INFO - Batch 998/1250\n", "2025-09-24 13:38:28,029 - INFO - Batch 999/1250\n", "2025-09-24 13:38:28,092 - INFO - Batch 1000/1250\n", "2025-09-24 13:38:28,156 - INFO - Batch 1001/1250\n", "2025-09-24 13:38:28,217 - INFO - Batch 1002/1250\n", "2025-09-24 13:38:28,284 - INFO - Batch 1003/1250\n", "2025-09-24 13:38:28,343 - INFO - Batch 1004/1250\n", "2025-09-24 13:38:28,406 - INFO - Batch 1005/1250\n", "2025-09-24 13:38:28,467 - INFO - Batch 1006/1250\n", "2025-09-24 13:38:28,533 - INFO - Batch 1007/1250\n", "2025-09-24 13:38:28,597 - INFO - Batch 1008/1250\n", "2025-09-24 13:38:28,664 - INFO - Batch 1009/1250\n", "2025-09-24 13:38:28,730 - INFO - Batch 1010/1250\n", "2025-09-24 13:38:28,789 - INFO - Batch 1011/1250\n", "2025-09-24 13:38:28,852 - INFO - Batch 1012/1250\n", "2025-09-24 13:38:28,912 - INFO - Batch 1013/1250\n", "2025-09-24 13:38:28,977 - INFO - Batch 1014/1250\n", "2025-09-24 13:38:29,039 - INFO - Batch 1015/1250\n", "2025-09-24 13:38:29,107 - INFO - Batch 1016/1250\n", "2025-09-24 13:38:29,172 - INFO - Batch 1017/1250\n", "2025-09-24 13:38:29,241 - INFO - Batch 1018/1250\n", "2025-09-24 13:38:29,304 - INFO - Batch 1019/1250\n", "2025-09-24 13:38:29,364 - INFO - Batch 1020/1250\n", "2025-09-24 13:38:29,425 - INFO - Batch 1021/1250\n", "2025-09-24 13:38:29,489 - INFO - Batch 1022/1250\n", "2025-09-24 13:38:29,553 - INFO - Batch 1023/1250\n", "2025-09-24 13:38:29,615 - INFO - Batch 1024/1250\n", "2025-09-24 13:38:29,737 - INFO - Batch 1025/1250\n", "2025-09-24 13:38:29,805 - INFO - Batch 1026/1250\n", "2025-09-24 13:38:29,868 - INFO - Batch 1027/1250\n", "2025-09-24 13:38:29,929 - INFO - Batch 1028/1250\n", "2025-09-24 13:38:29,993 - INFO - Batch 1029/1250\n", "2025-09-24 13:38:30,056 - INFO - Batch 1030/1250\n", "2025-09-24 13:38:30,119 - INFO - Batch 1031/1250\n", "2025-09-24 13:38:30,183 - INFO - Batch 1032/1250\n", "2025-09-24 13:38:30,254 - INFO - Batch 1033/1250\n", "2025-09-24 13:38:30,318 - INFO - Batch 1034/1250\n", "2025-09-24 13:38:30,381 - INFO - Batch 1035/1250\n", "2025-09-24 13:38:30,447 - INFO - Batch 1036/1250\n", "2025-09-24 13:38:30,513 - INFO - Batch 1037/1250\n", "2025-09-24 13:38:30,642 - INFO - Batch 1038/1250\n", "2025-09-24 13:38:30,715 - INFO - Batch 1039/1250\n", "2025-09-24 13:38:30,782 - INFO - Batch 1040/1250\n", "2025-09-24 13:38:30,845 - INFO - Batch 1041/1250\n", "2025-09-24 13:38:30,906 - INFO - Batch 1042/1250\n", "2025-09-24 13:38:30,968 - INFO - Batch 1043/1250\n", "2025-09-24 13:38:31,030 - INFO - Batch 1044/1250\n", "2025-09-24 13:38:31,096 - INFO - Batch 1045/1250\n", "2025-09-24 13:38:31,158 - INFO - Batch 1046/1250\n", "2025-09-24 13:38:31,217 - INFO - Batch 1047/1250\n", "2025-09-24 13:38:31,281 - INFO - Batch 1048/1250\n", "2025-09-24 13:38:31,344 - INFO - Batch 1049/1250\n", "2025-09-24 13:38:31,405 - INFO - Batch 1050/1250\n", "2025-09-24 13:38:31,464 - INFO - Batch 1051/1250\n", "2025-09-24 13:38:31,532 - INFO - Batch 1052/1250\n", "2025-09-24 13:38:31,598 - INFO - Batch 1053/1250\n", "2025-09-24 13:38:31,662 - INFO - Batch 1054/1250\n", "2025-09-24 13:38:31,731 - INFO - Batch 1055/1250\n", "2025-09-24 13:38:31,794 - INFO - Batch 1056/1250\n", "2025-09-24 13:38:31,854 - INFO - Batch 1057/1250\n", "2025-09-24 13:38:31,920 - INFO - Batch 1058/1250\n", "2025-09-24 13:38:31,985 - INFO - Batch 1059/1250\n", "2025-09-24 13:38:32,047 - INFO - Batch 1060/1250\n", "2025-09-24 13:38:32,108 - INFO - Batch 1061/1250\n", "2025-09-24 13:38:32,169 - INFO - Batch 1062/1250\n", "2025-09-24 13:38:32,232 - INFO - Batch 1063/1250\n", "2025-09-24 13:38:32,296 - INFO - Batch 1064/1250\n", "2025-09-24 13:38:32,357 - INFO - Batch 1065/1250\n", "2025-09-24 13:38:32,422 - INFO - Batch 1066/1250\n", "2025-09-24 13:38:32,488 - INFO - Batch 1067/1250\n", "2025-09-24 13:38:32,552 - INFO - Batch 1068/1250\n", "2025-09-24 13:38:32,617 - INFO - Batch 1069/1250\n", "2025-09-24 13:38:32,679 - INFO - Batch 1070/1250\n", "2025-09-24 13:38:32,749 - INFO - Batch 1071/1250\n", "2025-09-24 13:38:32,811 - INFO - Batch 1072/1250\n", "2025-09-24 13:38:32,875 - INFO - Batch 1073/1250\n", "2025-09-24 13:38:32,936 - INFO - Batch 1074/1250\n", "2025-09-24 13:38:32,995 - INFO - Batch 1075/1250\n", "2025-09-24 13:38:33,061 - INFO - Batch 1076/1250\n", "2025-09-24 13:38:33,126 - INFO - Batch 1077/1250\n", "2025-09-24 13:38:33,189 - INFO - Batch 1078/1250\n", "2025-09-24 13:38:33,249 - INFO - Batch 1079/1250\n", "2025-09-24 13:38:33,315 - INFO - Batch 1080/1250\n", "2025-09-24 13:38:33,377 - INFO - Batch 1081/1250\n", "2025-09-24 13:38:33,437 - INFO - Batch 1082/1250\n", "2025-09-24 13:38:33,503 - INFO - Batch 1083/1250\n", "2025-09-24 13:38:33,569 - INFO - Batch 1084/1250\n", "2025-09-24 13:38:33,628 - INFO - Batch 1085/1250\n", "2025-09-24 13:38:33,708 - INFO - Batch 1086/1250\n", "2025-09-24 13:38:33,780 - INFO - Batch 1087/1250\n", "2025-09-24 13:38:33,842 - INFO - Batch 1088/1250\n", "2025-09-24 13:38:33,905 - INFO - Batch 1089/1250\n", "2025-09-24 13:38:33,963 - INFO - Batch 1090/1250\n", "2025-09-24 13:38:34,025 - INFO - Batch 1091/1250\n", "2025-09-24 13:38:34,085 - INFO - Batch 1092/1250\n", "2025-09-24 13:38:34,149 - INFO - Batch 1093/1250\n", "2025-09-24 13:38:34,212 - INFO - Batch 1094/1250\n", "2025-09-24 13:38:34,285 - INFO - Batch 1095/1250\n", "2025-09-24 13:38:34,348 - INFO - Batch 1096/1250\n", "2025-09-24 13:38:34,412 - INFO - Batch 1097/1250\n", "2025-09-24 13:38:34,477 - INFO - Batch 1098/1250\n", "2025-09-24 13:38:34,537 - INFO - Batch 1099/1250\n", "2025-09-24 13:38:34,607 - INFO - Batch 1100/1250\n", "2025-09-24 13:38:34,674 - INFO - Batch 1101/1250\n", "2025-09-24 13:38:34,738 - INFO - Batch 1102/1250\n", "2025-09-24 13:38:34,800 - INFO - Batch 1103/1250\n", "2025-09-24 13:38:34,863 - INFO - Batch 1104/1250\n", "2025-09-24 13:38:34,932 - INFO - Batch 1105/1250\n", "2025-09-24 13:38:34,997 - INFO - Batch 1106/1250\n", "2025-09-24 13:38:35,062 - INFO - Batch 1107/1250\n", "2025-09-24 13:38:35,125 - INFO - Batch 1108/1250\n", "2025-09-24 13:38:35,189 - INFO - Batch 1109/1250\n", "2025-09-24 13:38:35,256 - INFO - Batch 1110/1250\n", "2025-09-24 13:38:35,318 - INFO - Batch 1111/1250\n", "2025-09-24 13:38:35,382 - INFO - Batch 1112/1250\n", "2025-09-24 13:38:35,444 - INFO - Batch 1113/1250\n", "2025-09-24 13:38:35,510 - INFO - Batch 1114/1250\n", "2025-09-24 13:38:35,570 - INFO - Batch 1115/1250\n", "2025-09-24 13:38:35,634 - INFO - Batch 1116/1250\n", "2025-09-24 13:38:35,698 - INFO - Batch 1117/1250\n", "2025-09-24 13:38:35,760 - INFO - Batch 1118/1250\n", "2025-09-24 13:38:35,823 - INFO - Batch 1119/1250\n", "2025-09-24 13:38:35,882 - INFO - Batch 1120/1250\n", "2025-09-24 13:38:35,948 - INFO - Batch 1121/1250\n", "2025-09-24 13:38:36,013 - INFO - Batch 1122/1250\n", "2025-09-24 13:38:36,077 - INFO - Batch 1123/1250\n", "2025-09-24 13:38:36,142 - INFO - Batch 1124/1250\n", "2025-09-24 13:38:36,202 - INFO - Batch 1125/1250\n", "2025-09-24 13:38:36,264 - INFO - Batch 1126/1250\n", "2025-09-24 13:38:36,324 - INFO - Batch 1127/1250\n", "2025-09-24 13:38:36,385 - INFO - Batch 1128/1250\n", "2025-09-24 13:38:36,453 - INFO - Batch 1129/1250\n", "2025-09-24 13:38:36,515 - INFO - Batch 1130/1250\n", "2025-09-24 13:38:36,578 - INFO - Batch 1131/1250\n", "2025-09-24 13:38:36,644 - INFO - Batch 1132/1250\n", "2025-09-24 13:38:36,706 - INFO - Batch 1133/1250\n", "2025-09-24 13:38:36,768 - INFO - Batch 1134/1250\n", "2025-09-24 13:38:36,831 - INFO - Batch 1135/1250\n", "2025-09-24 13:38:36,898 - INFO - Batch 1136/1250\n", "2025-09-24 13:38:36,959 - INFO - Batch 1137/1250\n", "2025-09-24 13:38:37,022 - INFO - Batch 1138/1250\n", "2025-09-24 13:38:37,085 - INFO - Batch 1139/1250\n", "2025-09-24 13:38:37,152 - INFO - Batch 1140/1250\n", "2025-09-24 13:38:37,217 - INFO - Batch 1141/1250\n", "2025-09-24 13:38:37,281 - INFO - Batch 1142/1250\n", "2025-09-24 13:38:37,343 - INFO - Batch 1143/1250\n", "2025-09-24 13:38:37,405 - INFO - Batch 1144/1250\n", "2025-09-24 13:38:37,469 - INFO - Batch 1145/1250\n", "2025-09-24 13:38:37,534 - INFO - Batch 1146/1250\n", "2025-09-24 13:38:37,596 - INFO - Batch 1147/1250\n", "2025-09-24 13:38:37,658 - INFO - Batch 1148/1250\n", "2025-09-24 13:38:37,722 - INFO - Batch 1149/1250\n", "2025-09-24 13:38:37,786 - INFO - Batch 1150/1250\n", "2025-09-24 13:38:37,851 - INFO - Batch 1151/1250\n", "2025-09-24 13:38:37,916 - INFO - Batch 1152/1250\n", "2025-09-24 13:38:37,984 - INFO - Batch 1153/1250\n", "2025-09-24 13:38:38,042 - INFO - Batch 1154/1250\n", "2025-09-24 13:38:38,105 - INFO - Batch 1155/1250\n", "2025-09-24 13:38:38,168 - INFO - Batch 1156/1250\n", "2025-09-24 13:38:38,241 - INFO - Batch 1157/1250\n", "2025-09-24 13:38:38,306 - INFO - Batch 1158/1250\n", "2025-09-24 13:38:38,370 - INFO - Batch 1159/1250\n", "2025-09-24 13:38:38,440 - INFO - Batch 1160/1250\n", "2025-09-24 13:38:38,503 - INFO - Batch 1161/1250\n", "2025-09-24 13:38:38,567 - INFO - Batch 1162/1250\n", "2025-09-24 13:38:38,635 - INFO - Batch 1163/1250\n", "2025-09-24 13:38:38,699 - INFO - Batch 1164/1250\n", "2025-09-24 13:38:38,771 - INFO - Batch 1165/1250\n", "2025-09-24 13:38:38,845 - INFO - Batch 1166/1250\n", "2025-09-24 13:38:38,929 - INFO - Batch 1167/1250\n", "2025-09-24 13:38:39,013 - INFO - Batch 1168/1250\n", "2025-09-24 13:38:39,093 - INFO - Batch 1169/1250\n", "2025-09-24 13:38:39,178 - INFO - Batch 1170/1250\n", "2025-09-24 13:38:39,252 - INFO - Batch 1171/1250\n", "2025-09-24 13:38:39,323 - INFO - Batch 1172/1250\n", "2025-09-24 13:38:39,410 - INFO - Batch 1173/1250\n", "2025-09-24 13:38:39,529 - INFO - Batch 1174/1250\n", "2025-09-24 13:38:39,608 - INFO - Batch 1175/1250\n", "2025-09-24 13:38:39,683 - INFO - Batch 1176/1250\n", "2025-09-24 13:38:39,758 - INFO - Batch 1177/1250\n", "2025-09-24 13:38:39,833 - INFO - Batch 1178/1250\n", "2025-09-24 13:38:39,907 - INFO - Batch 1179/1250\n", "2025-09-24 13:38:39,979 - INFO - Batch 1180/1250\n", "2025-09-24 13:38:40,050 - INFO - Batch 1181/1250\n", "2025-09-24 13:38:40,123 - INFO - Batch 1182/1250\n", "2025-09-24 13:38:40,194 - INFO - Batch 1183/1250\n", "2025-09-24 13:38:40,268 - INFO - Batch 1184/1250\n", "2025-09-24 13:38:40,341 - INFO - Batch 1185/1250\n", "2025-09-24 13:38:40,413 - INFO - Batch 1186/1250\n", "2025-09-24 13:38:40,488 - INFO - Batch 1187/1250\n", "2025-09-24 13:38:40,561 - INFO - Batch 1188/1250\n", "2025-09-24 13:38:40,633 - INFO - Batch 1189/1250\n", "2025-09-24 13:38:40,706 - INFO - Batch 1190/1250\n", "2025-09-24 13:38:40,776 - INFO - Batch 1191/1250\n", "2025-09-24 13:38:40,850 - INFO - Batch 1192/1250\n", "2025-09-24 13:38:40,933 - INFO - Batch 1193/1250\n", "2025-09-24 13:38:41,007 - INFO - Batch 1194/1250\n", "2025-09-24 13:38:41,079 - INFO - Batch 1195/1250\n", "2025-09-24 13:38:41,150 - INFO - Batch 1196/1250\n", "2025-09-24 13:38:41,217 - INFO - Batch 1197/1250\n", "2025-09-24 13:38:41,290 - INFO - Batch 1198/1250\n", "2025-09-24 13:38:41,364 - INFO - Batch 1199/1250\n", "2025-09-24 13:38:41,431 - INFO - Batch 1200/1250\n", "2025-09-24 13:38:41,496 - INFO - Batch 1201/1250\n", "2025-09-24 13:38:41,562 - INFO - Batch 1202/1250\n", "2025-09-24 13:38:41,626 - INFO - Batch 1203/1250\n", "2025-09-24 13:38:41,691 - INFO - Batch 1204/1250\n", "2025-09-24 13:38:41,760 - INFO - Batch 1205/1250\n", "2025-09-24 13:38:41,821 - INFO - Batch 1206/1250\n", "2025-09-24 13:38:41,886 - INFO - Batch 1207/1250\n", "2025-09-24 13:38:41,954 - INFO - Batch 1208/1250\n", "2025-09-24 13:38:42,036 - INFO - Batch 1209/1250\n", "2025-09-24 13:38:42,104 - INFO - Batch 1210/1250\n", "2025-09-24 13:38:42,165 - INFO - Batch 1211/1250\n", "2025-09-24 13:38:42,232 - INFO - Batch 1212/1250\n", "2025-09-24 13:38:42,295 - INFO - Batch 1213/1250\n", "2025-09-24 13:38:42,359 - INFO - Batch 1214/1250\n", "2025-09-24 13:38:42,422 - INFO - Batch 1215/1250\n", "2025-09-24 13:38:42,486 - INFO - Batch 1216/1250\n", "2025-09-24 13:38:42,550 - INFO - Batch 1217/1250\n", "2025-09-24 13:38:42,615 - INFO - Batch 1218/1250\n", "2025-09-24 13:38:42,678 - INFO - Batch 1219/1250\n", "2025-09-24 13:38:42,741 - INFO - Batch 1220/1250\n", "2025-09-24 13:38:42,809 - INFO - Batch 1221/1250\n", "2025-09-24 13:38:42,871 - INFO - Batch 1222/1250\n", "2025-09-24 13:38:42,932 - INFO - Batch 1223/1250\n", "2025-09-24 13:38:42,994 - INFO - Batch 1224/1250\n", "2025-09-24 13:38:43,057 - INFO - Batch 1225/1250\n", "2025-09-24 13:38:43,116 - INFO - Batch 1226/1250\n", "2025-09-24 13:38:43,176 - INFO - Batch 1227/1250\n", "2025-09-24 13:38:43,236 - INFO - Batch 1228/1250\n", "2025-09-24 13:38:43,296 - INFO - Batch 1229/1250\n", "2025-09-24 13:38:43,362 - INFO - Batch 1230/1250\n", "2025-09-24 13:38:43,424 - INFO - Batch 1231/1250\n", "2025-09-24 13:38:43,489 - INFO - Batch 1232/1250\n", "2025-09-24 13:38:43,557 - INFO - Batch 1233/1250\n", "2025-09-24 13:38:43,623 - INFO - Batch 1234/1250\n", "2025-09-24 13:38:43,687 - INFO - Batch 1235/1250\n", "2025-09-24 13:38:43,748 - INFO - Batch 1236/1250\n", "2025-09-24 13:38:43,813 - INFO - Batch 1237/1250\n", "2025-09-24 13:38:43,875 - INFO - Batch 1238/1250\n", "2025-09-24 13:38:43,936 - INFO - Batch 1239/1250\n", "2025-09-24 13:38:43,995 - INFO - Batch 1240/1250\n", "2025-09-24 13:38:44,060 - INFO - Batch 1241/1250\n", "2025-09-24 13:38:44,124 - INFO - Batch 1242/1250\n", "2025-09-24 13:38:44,187 - INFO - Batch 1243/1250\n", "2025-09-24 13:38:44,252 - INFO - Batch 1244/1250\n", "2025-09-24 13:38:44,315 - INFO - Batch 1245/1250\n", "2025-09-24 13:38:44,378 - INFO - Batch 1246/1250\n", "2025-09-24 13:38:44,441 - INFO - Batch 1247/1250\n", "2025-09-24 13:38:44,503 - INFO - Batch 1248/1250\n", "2025-09-24 13:38:44,566 - INFO - Batch 1249/1250\n", "2025-09-24 13:38:44,634 - INFO - Batch 1250/1250\n", "2025-09-24 13:38:44,696 - INFO - Took 81.66 seconds\n", "2025-09-24 13:38:44,699 - INFO - Generated embeddings with dimensions torch.Size([10000, 512])\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Embeddings generated successfully!\n", "Embedding shape: torch.Size([10000, 512])\n", "Embedding dimensions: 512\n" ] } ], "source": [ "# Generate embeddings using antiBERTy model\n", "print(\"Generating embeddings using antiBERTy model...\")\n", "\n", "# First, ensure the data has the required duplicate_count column\n", "if 'duplicate_count' not in heavy_sample.columns:\n", " heavy_sample = heavy_sample.copy()\n", " heavy_sample['duplicate_count'] = 1\n", " print(\"Added duplicate_count column for amulety compatibility\")\n", "\n", "embeddings, data = embed_airr(\n", " airr=heavy_sample,\n", " chain=\"H\", # Heavy chain only\n", " model=\"antiberty\", # antiBERTy - Antibody BERT model\n", " sequence_col=\"sequence_vdj_aa\",\n", " cell_id_col=\"cell_id\",\n", " batch_size=8, # Adjust based on available memory\n", " output_type=\"pickle\" \n", ")\n", "\n", "print(f\"Embeddings generated successfully!\")\n", "print(f\"Embedding shape: {embeddings.shape}\")\n", "print(f\"Embedding dimensions: {embeddings.shape[1]}\")\n", "\n", "# Convert to numpy for downstream analysis\n", "X_embeddings = embeddings.numpy() if hasattr(embeddings, 'numpy') else embeddings" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Prepare Data for Machine Learning\n", "\n", "We'll prepare the data for downstream machine learning tasks. In this example, we'll predict V gene family usage from the embeddings." ] }, { "cell_type": "code", "execution_count": 65, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Data after removing missing labels:\n", "Samples: 10000\n", "Features: 512\n", "Unique subjects: 73\n", "\n", "Class distribution:\n", " IGHV3: 4755\n", " IGHV4: 2271\n", " IGHV1: 1968\n", " IGHV5: 450\n", " IGHV2: 421\n", " IGHV6: 87\n", " IGHV7: 46\n", " IGHV4/OR15: 1\n", " IGHV3/OR16: 1\n" ] } ], "source": [ "# Prepare labels for prediction task\n", "# Example: Predict V gene family\n", "target_column = 'v_call_family'\n", "y_labels = heavy_sample[target_column].values\n", "subjects = heavy_sample['subject'].values\n", "\n", "# Remove samples with missing labels\n", "valid_mask = pd.notna(y_labels)\n", "X_embeddings = X_embeddings[valid_mask]\n", "y_labels = y_labels[valid_mask]\n", "subjects = subjects[valid_mask]\n", "\n", "print(f\"Data after removing missing labels:\")\n", "print(f\"Samples: {len(y_labels)}\")\n", "print(f\"Features: {X_embeddings.shape[1]}\")\n", "print(f\"Unique subjects: {len(np.unique(subjects))}\")\n", "\n", "# Check class distribution\n", "class_counts = Counter(y_labels)\n", "print(f\"\\nClass distribution:\")\n", "for class_name, count in class_counts.most_common():\n", " print(f\" {class_name}: {count}\")" ] }, { "cell_type": "code", "execution_count": 66, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "After filtering classes with < 200 samples:\n", "Samples: 9865\n", "Classes: 5\n", "Valid classes: ['IGHV3', 'IGHV4', 'IGHV5', 'IGHV1', 'IGHV2']\n" ] } ], "source": [ "# Filter classes with sufficient samples\n", "MIN_CLASS_SIZE = 200 # Minimum samples per class\n", "MAX_CLASS_SIZE = 2000 # Maximum samples per class for balanced training\n", "\n", "# Keep only classes with sufficient samples\n", "valid_classes = [cls for cls, count in class_counts.items() if count >= MIN_CLASS_SIZE]\n", "valid_mask = np.isin(y_labels, valid_classes)\n", "\n", "X_filtered = X_embeddings[valid_mask]\n", "y_filtered = y_labels[valid_mask]\n", "subjects_filtered = subjects[valid_mask]\n", "\n", "print(f\"After filtering classes with < {MIN_CLASS_SIZE} samples:\")\n", "print(f\"Samples: {len(y_filtered)}\")\n", "print(f\"Classes: {len(valid_classes)}\")\n", "print(f\"Valid classes: {valid_classes}\")" ] }, { "cell_type": "code", "execution_count": 68, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Final dataset after subsampling:\n", "Samples: 6839\n", "Features: 512\n", "Classes: 5\n", "\n", "Final class distribution:\n", " IGHV3: 2000\n", " IGHV4: 2000\n", " IGHV1: 1968\n", " IGHV5: 450\n", " IGHV2: 421\n" ] } ], "source": [ "# Subsample large classes for balanced training\n", "def subsample_classes(X, y, subjects, max_size=MAX_CLASS_SIZE, random_state=42):\n", " \"\"\"Subsample classes to have at most max_size samples each\"\"\"\n", " np.random.seed(random_state)\n", " \n", " indices_to_keep = []\n", " for class_name in np.unique(y):\n", " class_indices = np.where(y == class_name)[0]\n", " if len(class_indices) <= max_size:\n", " indices_to_keep.extend(class_indices)\n", " else:\n", " # Randomly sample max_size indices\n", " sampled_indices = np.random.choice(class_indices, size=max_size, replace=False)\n", " indices_to_keep.extend(sampled_indices)\n", " \n", " indices_to_keep = np.array(indices_to_keep)\n", " return X[indices_to_keep], y[indices_to_keep], subjects[indices_to_keep]\n", "\n", "# Apply subsampling\n", "X_final, y_final, subjects_final = subsample_classes(X_filtered, y_filtered, subjects_filtered)\n", "\n", "print(f\"Final dataset after subsampling:\")\n", "print(f\"Samples: {len(y_final)}\")\n", "print(f\"Features: {X_final.shape[1]}\")\n", "print(f\"Classes: {len(np.unique(y_final))}\")\n", "\n", "# Final class distribution\n", "final_class_counts = Counter(y_final)\n", "print(f\"\\nFinal class distribution:\")\n", "for class_name, count in final_class_counts.most_common():\n", " print(f\" {class_name}: {count}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Machine Learning Pipeline\n", "\n", "We'll implement a robust machine learning pipeline using nested cross-validation to evaluate the predictive power of our embeddings." ] }, { "cell_type": "code", "execution_count": 69, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Setting up nested cross-validation:\n", "Outer folds: 5\n", "Inner folds: 3\n", "Hyperparameter grid: {'C': [0.01, 0.1, 1, 10, 100]}\n" ] } ], "source": [ "# Set up nested cross-validation\n", "n_splits_outer = 5\n", "n_splits_inner = 3\n", "\n", "# Use StratifiedGroupKFold to ensure subjects don't appear in both train and test\n", "outer_cv = StratifiedGroupKFold(n_splits=n_splits_outer, shuffle=True, random_state=42)\n", "inner_cv = StratifiedGroupKFold(n_splits=n_splits_inner, shuffle=True, random_state=43)\n", "\n", "# Hyperparameter grid for SVM\n", "param_grid = {\n", " \"C\": [0.01, 0.1, 1, 10, 100]\n", "}\n", "\n", "print(f\"Setting up nested cross-validation:\")\n", "print(f\"Outer folds: {n_splits_outer}\")\n", "print(f\"Inner folds: {n_splits_inner}\")\n", "print(f\"Hyperparameter grid: {param_grid}\")" ] }, { "cell_type": "code", "execution_count": 70, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Running nested cross-validation...\n", "This may take a few minutes depending on dataset size and computational resources.\n", "\n", "--- Outer Fold 1 ---\n", "Train size: 6211, Test size: 628\n", "Best C: 10\n", "F1 Score: 1.0000\n", "MCC Score: 1.0000\n", "Accuracy: 1.0000\n", "\n", "--- Outer Fold 2 ---\n", "Train size: 5501, Test size: 1338\n", "Best C: 10\n", "F1 Score: 0.9978\n", "MCC Score: 0.9970\n", "Accuracy: 0.9978\n", "\n", "--- Outer Fold 3 ---\n", "Train size: 5795, Test size: 1044\n", "Best C: 1\n", "F1 Score: 0.9990\n", "MCC Score: 0.9987\n", "Accuracy: 0.9990\n", "\n", "--- Outer Fold 4 ---\n", "Train size: 3993, Test size: 2846\n", "Best C: 0.1\n", "F1 Score: 0.9961\n", "MCC Score: 0.9947\n", "Accuracy: 0.9961\n", "\n", "--- Outer Fold 5 ---\n", "Train size: 5856, Test size: 983\n", "Best C: 0.1\n", "F1 Score: 0.9980\n", "MCC Score: 0.9972\n", "Accuracy: 0.9980\n", "\n", "Nested cross-validation completed!\n" ] } ], "source": [ "# Run nested cross-validation\n", "print(\"Running nested cross-validation...\")\n", "print(\"This may take a few minutes depending on dataset size and computational resources.\")\n", "\n", "# Initialize results storage\n", "cv_results = {\n", " 'fold': [],\n", " 'f1_score': [],\n", " 'mcc_score': [],\n", " 'accuracy': [],\n", " 'best_C': []\n", "}\n", "\n", "# Perform nested cross-validation\n", "fold_idx = 1\n", "for train_idx, test_idx in outer_cv.split(X_final, y_final, subjects_final):\n", " print(f\"\\n--- Outer Fold {fold_idx} ---\")\n", " \n", " # Split data\n", " X_train, X_test = X_final[train_idx], X_final[test_idx]\n", " y_train, y_test = y_final[train_idx], y_final[test_idx]\n", " subjects_train = subjects_final[train_idx]\n", " \n", " print(f\"Train size: {len(X_train)}, Test size: {len(X_test)}\")\n", " \n", " # Inner cross-validation for hyperparameter tuning\n", " svm_model = SVC(kernel=\"rbf\", class_weight=\"balanced\", random_state=42)\n", " \n", " grid_search = GridSearchCV(\n", " estimator=svm_model,\n", " param_grid=param_grid,\n", " cv=inner_cv,\n", " scoring=\"f1_weighted\",\n", " n_jobs=-1\n", " )\n", " \n", " # Fit grid search\n", " grid_search.fit(X_train, y_train, groups=subjects_train)\n", " \n", " # Get best model and make predictions\n", " best_model = grid_search.best_estimator_\n", " y_pred = best_model.predict(X_test)\n", " \n", " # Calculate metrics\n", " f1 = f1_score(y_test, y_pred, average=\"weighted\")\n", " mcc = matthews_corrcoef(y_test, y_pred)\n", " acc = accuracy_score(y_test, y_pred)\n", " \n", " # Store results\n", " cv_results['fold'].append(fold_idx)\n", " cv_results['f1_score'].append(f1)\n", " cv_results['mcc_score'].append(mcc)\n", " cv_results['accuracy'].append(acc)\n", " cv_results['best_C'].append(grid_search.best_params_['C'])\n", " \n", " print(f\"Best C: {grid_search.best_params_['C']}\")\n", " print(f\"F1 Score: {f1:.4f}\")\n", " print(f\"MCC Score: {mcc:.4f}\")\n", " print(f\"Accuracy: {acc:.4f}\")\n", " \n", " fold_idx += 1\n", "\n", "print(\"\\nNested cross-validation completed!\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Results Analysis\n", "\n", "Let's analyze the performance of our model and visualize the results." ] }, { "cell_type": "code", "execution_count": 71, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Cross-Validation Results Summary:\n", "========================================\n", "F1_SCORE: 0.9982 ± 0.0015\n", "MCC_SCORE: 0.9975 ± 0.0020\n", "ACCURACY: 0.9982 ± 0.0015\n", "\n", "Best hyperparameters (C values): [10.0, 10.0, 1.0, 0.1, 0.1]\n", "Most common C value: 10.0\n", "\n", "Detailed Results by Fold:\n", " fold f1_score mcc_score accuracy best_C\n", "0 1 1.0000 1.0000 1.0000 10.0\n", "1 2 0.9978 0.9970 0.9978 10.0\n", "2 3 0.9990 0.9987 0.9990 1.0\n", "3 4 0.9961 0.9947 0.9961 0.1\n", "4 5 0.9980 0.9972 0.9980 0.1\n" ] } ], "source": [ "# Convert results to DataFrame\n", "results_df = pd.DataFrame(cv_results)\n", "\n", "# Calculate summary statistics\n", "print(\"Cross-Validation Results Summary:\")\n", "print(\"=\" * 40)\n", "\n", "for metric in ['f1_score', 'mcc_score', 'accuracy']:\n", " mean_score = results_df[metric].mean()\n", " std_score = results_df[metric].std()\n", " print(f\"{metric.upper()}: {mean_score:.4f} ± {std_score:.4f}\")\n", "\n", "print(f\"\\nBest hyperparameters (C values): {results_df['best_C'].tolist()}\")\n", "print(f\"Most common C value: {Counter(results_df['best_C']).most_common(1)[0][0]}\")\n", "\n", "# Display detailed results\n", "print(\"\\nDetailed Results by Fold:\")\n", "print(results_df.round(4))" ] }, { "cell_type": "code", "execution_count": 72, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Visualize results\n", "fig, axes = plt.subplots(1, 3, figsize=(15, 5))\n", "\n", "metrics = ['f1_score', 'mcc_score', 'accuracy']\n", "metric_names = ['F1 Score', 'MCC Score', 'Accuracy']\n", "\n", "for i, (metric, name) in enumerate(zip(metrics, metric_names)):\n", " axes[i].bar(results_df['fold'], results_df[metric], alpha=0.7, color=f'C{i}')\n", " axes[i].axhline(y=results_df[metric].mean(), color='red', linestyle='--', \n", " label=f'Mean: {results_df[metric].mean():.3f}')\n", " axes[i].set_xlabel('Fold')\n", " axes[i].set_ylabel(name)\n", " axes[i].set_title(f'{name} by Fold')\n", " axes[i].legend()\n", " axes[i].set_ylim(0, 1)\n", "\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Model Comparison with Random Baseline\n", "\n", "To validate that our embeddings contain meaningful information, let's compare against a random baseline." ] }, { "cell_type": "code", "execution_count": 76, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Running random baseline comparison...\n", "Random baseline - Fold 1\n", "Random baseline - Fold 2\n", "Random baseline - Fold 3\n", "Random baseline - Fold 4\n", "Random baseline - Fold 5\n", "Random baseline completed!\n" ] } ], "source": [ "# Create random baseline by shuffling embeddings\n", "print(\"Running random baseline comparison...\")\n", "\n", "# Shuffle the embeddings while keeping labels intact\n", "# np.random.seed(42)\n", "X_random = X_final.copy()\n", "# Shuffle both rows and columns to completely randomize\n", "X_random = X_random[np.random.permutation(X_random.shape[0])]\n", "X_random = X_random[:, np.random.permutation(X_random.shape[1])]\n", "\n", "# Run the same cross-validation with random embeddings\n", "random_results = {\n", " 'fold': [],\n", " 'f1_score': [],\n", " 'mcc_score': [],\n", " 'accuracy': []\n", "}\n", "\n", "fold_idx = 1\n", "for train_idx, test_idx in outer_cv.split(X_random, y_final, subjects_final):\n", " print(f\"Random baseline - Fold {fold_idx}\")\n", " \n", " X_train, X_test = X_random[train_idx], X_random[test_idx]\n", " y_train, y_test = y_final[train_idx], y_final[test_idx]\n", " subjects_train = subjects_final[train_idx]\n", " \n", " # Use simple SVM with default parameters for speed\n", " svm_model = SVC(kernel=\"rbf\", class_weight=\"balanced\", C=1.0, random_state=42)\n", " svm_model.fit(X_train, y_train)\n", " y_pred = svm_model.predict(X_test)\n", " \n", " # Calculate metrics\n", " f1 = f1_score(y_test, y_pred, average=\"weighted\")\n", " mcc = matthews_corrcoef(y_test, y_pred)\n", " acc = accuracy_score(y_test, y_pred)\n", " \n", " random_results['fold'].append(fold_idx)\n", " random_results['f1_score'].append(f1)\n", " random_results['mcc_score'].append(mcc)\n", " random_results['accuracy'].append(acc)\n", " \n", " fold_idx += 1\n", "\n", "random_df = pd.DataFrame(random_results)\n", "print(\"Random baseline completed!\")" ] }, { "cell_type": "code", "execution_count": 77, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Performance Comparison: Amulety Embeddings vs Random Baseline\n", "============================================================\n", "F1_SCORE:\n", " Amulety Embeddings: 0.9982 ± 0.0015\n", " Random Baseline: 0.1981 ± 0.0061\n", " Improvement: 404.0%\n", "\n", "MCC_SCORE:\n", " Amulety Embeddings: 0.9975 ± 0.0020\n", " Random Baseline: 0.0019 ± 0.0119\n", " Improvement: 51836.2%\n", "\n", "ACCURACY:\n", " Amulety Embeddings: 0.9982 ± 0.0015\n", " Random Baseline: 0.1783 ± 0.0093\n", " Improvement: 459.9%\n", "\n", "Summary Table:\n", " Metric Amulety Random Improvement\n", " F1_SCORE 0.9982 ± 0.0015 0.1981 ± 0.0061 404.0%\n", "MCC_SCORE 0.9975 ± 0.0020 0.0019 ± 0.0119 51836.2%\n", " ACCURACY 0.9982 ± 0.0015 0.1783 ± 0.0093 459.9%\n" ] } ], "source": [ "# Compare results\n", "print(\"Performance Comparison: Amulety Embeddings vs Random Baseline\")\n", "print(\"=\" * 60)\n", "\n", "comparison_data = []\n", "for metric in ['f1_score', 'mcc_score', 'accuracy']:\n", " amulety_mean = results_df[metric].mean()\n", " amulety_std = results_df[metric].std()\n", " random_mean = random_df[metric].mean()\n", " random_std = random_df[metric].std()\n", " improvement = ((amulety_mean - random_mean) / random_mean) * 100\n", " \n", " comparison_data.append({\n", " 'Metric': metric.upper(),\n", " 'Amulety': f\"{amulety_mean:.4f} ± {amulety_std:.4f}\",\n", " 'Random': f\"{random_mean:.4f} ± {random_std:.4f}\",\n", " 'Improvement': f\"{improvement:.1f}%\"\n", " })\n", " \n", " print(f\"{metric.upper()}:\")\n", " print(f\" Amulety Embeddings: {amulety_mean:.4f} ± {amulety_std:.4f}\")\n", " print(f\" Random Baseline: {random_mean:.4f} ± {random_std:.4f}\")\n", " print(f\" Improvement: {improvement:.1f}%\")\n", " print()\n", "\n", "comparison_df = pd.DataFrame(comparison_data)\n", "print(\"Summary Table:\")\n", "print(comparison_df.to_string(index=False))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Visualization of Results" ] }, { "cell_type": "code", "execution_count": 79, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Create comparison visualization\n", "fig, ax = plt.subplots(1, 1, figsize=(10, 6))\n", "\n", "metrics = ['f1_score', 'mcc_score', 'accuracy']\n", "metric_names = ['F1 Score', 'MCC Score', 'Accuracy']\n", "\n", "x = np.arange(len(metrics))\n", "width = 0.35\n", "\n", "amulety_means = [results_df[metric].mean() for metric in metrics]\n", "amulety_stds = [results_df[metric].std() for metric in metrics]\n", "random_means = [random_df[metric].mean() for metric in metrics]\n", "random_stds = [random_df[metric].std() for metric in metrics]\n", "\n", "bars1 = ax.bar(x - width/2, amulety_means, width, yerr=amulety_stds, \n", " label='Amulety Embeddings', alpha=0.8, capsize=5)\n", "bars2 = ax.bar(x + width/2, random_means, width, yerr=random_stds, \n", " label='Random Baseline', alpha=0.8, capsize=5)\n", "\n", "ax.set_xlabel('Metrics')\n", "ax.set_ylabel('Score')\n", "# ax.set_title('Performance Comparison: Amulety Embeddings vs Random Baseline')\n", "ax.set_xticks(x)\n", "ax.set_xticklabels(metric_names)\n", "ax.legend()\n", "ax.set_ylim(0, 1)\n", "\n", "# Add value labels on bars\n", "def add_value_labels(bars, values, stds):\n", " for bar, val, std in zip(bars, values, stds):\n", " height = bar.get_height()\n", " ax.text(bar.get_x() + bar.get_width()/2., height + std + 0.01,\n", " f'{val:.3f}', ha='center', va='bottom', fontsize=9)\n", "\n", "add_value_labels(bars1, amulety_means, amulety_stds)\n", "add_value_labels(bars2, random_means, random_stds)\n", "\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Save results for future reference\n", "results_df.to_csv('ml_cv_results.csv', index=False)\n", "random_df.to_csv('random_baseline_results.csv', index=False)\n", "comparison_df.to_csv('performance_comparison.csv', index=False)" ] } ], "metadata": { "kernelspec": { "display_name": "torchen", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.13" } }, "nbformat": 4, "nbformat_minor": 4 }