{ "cells": [ { "cell_type": "code", "execution_count": 3, "id": "18b023f3-763d-4c3b-9ec4-628b03521d28", "metadata": {}, "outputs": [], "source": [ "import os\n", "from sklearn.model_selection import train_test_split\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", "\n", "\n", "import logging\n", "import sys\n", "import traceback\n", "from datetime import datetime\n", "\n", "from datasets import load_dataset\n", "\n", "from sentence_transformers import SentenceTransformer, losses\n", "from sentence_transformers.evaluation import EmbeddingSimilarityEvaluator,TripletEvaluator,SequentialEvaluator\n", "from sentence_transformers.similarity_functions import SimilarityFunction\n", "from sentence_transformers.trainer import SentenceTransformerTrainer\n", "from sentence_transformers.training_args import SentenceTransformerTrainingArguments\n", "import torch\n", "from sentence_transformers.training_args import BatchSamplers\n", "import pandas as pd\n", "from arabert.preprocess import ArabertPreprocessor\n", "from pathlib import Path" ] }, { "cell_type": "code", "execution_count": 13, "id": "d4cbdc18-633b-4967-8079-f789d98ad5f9", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "'\\nbased on me reading arabrtv02 doc i should preprocess my data beforer runing anything \\nto insert a space between numbers and characters and around punctuation characters.\\nexample below\\n###########\\nimport pandas as pd\\nfrom arabert.preprocess import ArabertPreprocessor\\nfrom pathlib import Path\\n\\nfrom arabert.preprocess import ArabertPreprocessor\\n\\nmodel_name=\"aubmindlab/bert-large-arabertv02\"\\narabert_prep = ArabertPreprocessor(model_name=model_name)\\n\\ntext = \"ولن نبالغ إذا قلنا: إن هاتف 7أو كمبيوتر المكتب في زمننا هذا ضروري\"\\narabert_prep.preprocess(text)\\n>>>>>>>>> output:\\'ولن نبالغ إذا قلنا : إن هاتف 7 أو كمبيوتر المكتب في زمننا هذا ضروري\\'\\n\\n'" ] }, "execution_count": 13, "metadata": {}, "output_type": "execute_result" } ], "source": [ "'''\n", "based on me reading arabrtv02 doc i should preprocess my data beforer runing anything \n", "to insert a space between numbers and characters and around punctuation characters.\n", "example below\n", "###########\n", "import pandas as pd\n", "from arabert.preprocess import ArabertPreprocessor\n", "from pathlib import Path\n", "\n", "from arabert.preprocess import ArabertPreprocessor\n", "\n", "model_name=\"aubmindlab/bert-large-arabertv02\"\n", "arabert_prep = ArabertPreprocessor(model_name=model_name)\n", "\n", "text = \"ولن نبالغ إذا قلنا: إن هاتف 7أو كمبيوتر المكتب في زمننا هذا ضروري\"\n", "arabert_prep.preprocess(text)\n", ">>>>>>>>> output:'ولن نبالغ إذا قلنا : إن هاتف 7 أو كمبيوتر المكتب في زمننا هذا ضروري'\n", "\n", "'''\n" ] }, { "cell_type": "code", "execution_count": 5, "id": "8d7ebfb5-e854-4f51-9b2c-408e79ed680e", "metadata": {}, "outputs": [ { "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", "
anchorpositivenegative
0ما هي المسألة الشرقية ؟المسألة الشرقية (بالإنجليزية: Eastern Question...وفي 14 نوفمبر 1990 م، وقعت الحكومة الألمانية م...
1كم تبعد بيت لحم على القدس؟بيت لحم (بالسريانية: ܒܝܬ ܠܚܡ ؛ باليونانية: Βηθ...كنائس بيت لحم تعتبر مدينة بيت لحم في فلسطين، م...
2متى ظهرت جماعة الاخوان المسلمين في سوريا؟الإخوان المسلمون في سوريا: أسسها طلاب في ثلاثي...ذكر يوسف ندا، مفوض العلاقات السياسية الدولية ل...
3من هو ريتشارد دوكينز؟كلينتون ريتشارد دوكينز عالمُ سلوك حيوان ، و عا...عام 2006، أسس دوكينز \"مؤسسة ريتشارد دوكينز للم...
4ما هي ضريبة الدخل ؟ضريبة الدخل هي ضريبة مباشرة تفرض على الأشخاص(س...الضريبة على الشركات  هي ضريبة مباشرة تفرضها ال...
............
5648879متى تجري الإنتخابات التشريعية في نيوزلندا؟متى تقوم الإنتخابات البرلمانية النيوزلندية؟متى وقعت غزوة تبوك؟
5648880هل يمكن علاج النمش عن طريق الجراحة؟كيف أتخلّص من النمش من خلال الجراحة؟ما هو علاج الصفار؟
5648881ما هو دعاء الاستخارة؟ما نص دعاء الاستخارة؟ما هو المشروع ؟
5648882كيف احضر عجينة البيتزا بالزبادي؟من طرق تحضير عجينة البيتزا بالزبادي؟من طرق تحضير البيتزا الإيطاليّة؟
5648883ما هي المساحة التي تشكلها قارة آسيا على الأرض؟ما المساحة التي تشكلها قارة آسيا على الأرض؟ما العوامل التي تؤثر على التفويض ؟
\n", "

5648884 rows × 3 columns

\n", "
" ], "text/plain": [ " anchor \\\n", "0 ما هي المسألة الشرقية ؟ \n", "1 كم تبعد بيت لحم على القدس؟ \n", "2 متى ظهرت جماعة الاخوان المسلمين في سوريا؟ \n", "3 من هو ريتشارد دوكينز؟ \n", "4 ما هي ضريبة الدخل ؟ \n", "... ... \n", "5648879 متى تجري الإنتخابات التشريعية في نيوزلندا؟ \n", "5648880 هل يمكن علاج النمش عن طريق الجراحة؟ \n", "5648881 ما هو دعاء الاستخارة؟ \n", "5648882 كيف احضر عجينة البيتزا بالزبادي؟ \n", "5648883 ما هي المساحة التي تشكلها قارة آسيا على الأرض؟ \n", "\n", " positive \\\n", "0 المسألة الشرقية (بالإنجليزية: Eastern Question... \n", "1 بيت لحم (بالسريانية: ܒܝܬ ܠܚܡ ؛ باليونانية: Βηθ... \n", "2 الإخوان المسلمون في سوريا: أسسها طلاب في ثلاثي... \n", "3 كلينتون ريتشارد دوكينز عالمُ سلوك حيوان ، و عا... \n", "4 ضريبة الدخل هي ضريبة مباشرة تفرض على الأشخاص(س... \n", "... ... \n", "5648879 متى تقوم الإنتخابات البرلمانية النيوزلندية؟ \n", "5648880 كيف أتخلّص من النمش من خلال الجراحة؟ \n", "5648881 ما نص دعاء الاستخارة؟ \n", "5648882 من طرق تحضير عجينة البيتزا بالزبادي؟ \n", "5648883 ما المساحة التي تشكلها قارة آسيا على الأرض؟ \n", "\n", " negative \n", "0 وفي 14 نوفمبر 1990 م، وقعت الحكومة الألمانية م... \n", "1 كنائس بيت لحم تعتبر مدينة بيت لحم في فلسطين، م... \n", "2 ذكر يوسف ندا، مفوض العلاقات السياسية الدولية ل... \n", "3 عام 2006، أسس دوكينز \"مؤسسة ريتشارد دوكينز للم... \n", "4 الضريبة على الشركات  هي ضريبة مباشرة تفرضها ال... \n", "... ... \n", "5648879 متى وقعت غزوة تبوك؟ \n", "5648880 ما هو علاج الصفار؟ \n", "5648881 ما هو المشروع ؟ \n", "5648882 من طرق تحضير البيتزا الإيطاليّة؟ \n", "5648883 ما العوامل التي تؤثر على التفويض ؟ \n", "\n", "[5648884 rows x 3 columns]" ] }, "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ "path = Path(r'/home/skiredj.abderrahman/khalil/sbert_training/semantic_textual_similarity_&_information_retrieval_dataset.csv')\n", "ds20 = pd.read_csv(path)\n", "ds20\n", "\n" ] }, { "cell_type": "code", "execution_count": 6, "id": "f9457b56-ebcc-4f66-be17-adf81d3471e5", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Preprocessing anchor...\n", "Preprocessing positive...\n", "Preprocessing negative...\n", "Preprocessing Complete!\n", " anchor \\\n", "0 ما هي المسألة الشرقية ؟ \n", "1 كم تبعد بيت لحم على القدس ؟ \n", "2 متى ظهرت جماعة الاخوان المسلمين في سوريا ؟ \n", "3 من هو ريتشارد دوكينز ؟ \n", "4 ما هي ضريبة الدخل ؟ \n", "\n", " positive \\\n", "0 المسألة الشرقية ( بالإنجليزية : Eastern Questi... \n", "1 بيت لحم ( بالسريانية : ؛ باليونانية : ؛ باللات... \n", "2 الإخوان المسلمون في سوريا : أسسها طلاب في ثلاث... \n", "3 كلينتون ريتشارد دوكينز عالم سلوك حيوان ، و عال... \n", "4 ضريبة الدخل هي ضريبة مباشرة تفرض على الأشخاص (... \n", "\n", " negative \n", "0 وفي 14 نوفمبر 1990 م ، وقعت الحكومة الألمانية ... \n", "1 كنائس بيت لحم تعتبر مدينة بيت لحم في فلسطين ، ... \n", "2 ذكر يوسف ندا ، مفوض العلاقات السياسية الدولية ... \n", "3 عام 2006 ، أسس دوكينز \" مؤسسة ريتشارد دوكينز ل... \n", "4 الضريبة على الشركات هي ضريبة مباشرة تفرضها الس... \n" ] } ], "source": [ "# 1. Initialize the Preprocessor\n", "model_name = \"bert-large-arabertv02\"\n", "arabert_prep = ArabertPreprocessor(model_name=model_name)\n", "\n", "# 2. Load your data\n", "path = Path(r'/home/skiredj.abderrahman/khalil/sbert_training/semantic_textual_similarity_&_information_retrieval_dataset.csv')\n", "\n", "ds20 = pd.read_csv(path)\n", "\n", "# 3. Apply the preprocessing to all three columns\n", "cols_to_clean = ['anchor', 'positive', 'negative']\n", "\n", "for col in cols_to_clean:\n", " print(f\"Preprocessing {col}...\")\n", " ds20[col] = ds20[col].apply(lambda x: arabert_prep.preprocess(str(x)))\n", "\n", "# 4. Check the results\n", "print(\"Preprocessing Complete!\")\n", "print(ds20.head())" ] }, { "cell_type": "code", "execution_count": 7, "id": "b5bf689b-1cce-4ac5-b1ce-accac544aa2e", "metadata": {}, "outputs": [ { "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", "
anchorpositivenegative
0ما هي المسألة الشرقية ؟المسألة الشرقية ( بالإنجليزية : Eastern Questi...وفي 14 نوفمبر 1990 م ، وقعت الحكومة الألمانية ...
1كم تبعد بيت لحم على القدس ؟بيت لحم ( بالسريانية : ؛ باليونانية : ؛ باللات...كنائس بيت لحم تعتبر مدينة بيت لحم في فلسطين ، ...
2متى ظهرت جماعة الاخوان المسلمين في سوريا ؟الإخوان المسلمون في سوريا : أسسها طلاب في ثلاث...ذكر يوسف ندا ، مفوض العلاقات السياسية الدولية ...
3من هو ريتشارد دوكينز ؟كلينتون ريتشارد دوكينز عالم سلوك حيوان ، و عال...عام 2006 ، أسس دوكينز \" مؤسسة ريتشارد دوكينز ل...
4ما هي ضريبة الدخل ؟ضريبة الدخل هي ضريبة مباشرة تفرض على الأشخاص (...الضريبة على الشركات هي ضريبة مباشرة تفرضها الس...
............
5648879متى تجري الإنتخابات التشريعية في نيوزلندا ؟متى تقوم الإنتخابات البرلمانية النيوزلندية ؟متى وقعت غزوة تبوك ؟
5648880هل يمكن علاج النمش عن طريق الجراحة ؟كيف أتخلص من النمش من خلال الجراحة ؟ما هو علاج الصفار ؟
5648881ما هو دعاء الاستخارة ؟ما نص دعاء الاستخارة ؟ما هو المشروع ؟
5648882كيف احضر عجينة البيتزا بالزبادي ؟من طرق تحضير عجينة البيتزا بالزبادي ؟من طرق تحضير البيتزا الإيطالية ؟
5648883ما هي المساحة التي تشكلها قارة آسيا على الأرض ؟ما المساحة التي تشكلها قارة آسيا على الأرض ؟ما العوامل التي تؤثر على التفويض ؟
\n", "

5648884 rows × 3 columns

\n", "
" ], "text/plain": [ " anchor \\\n", "0 ما هي المسألة الشرقية ؟ \n", "1 كم تبعد بيت لحم على القدس ؟ \n", "2 متى ظهرت جماعة الاخوان المسلمين في سوريا ؟ \n", "3 من هو ريتشارد دوكينز ؟ \n", "4 ما هي ضريبة الدخل ؟ \n", "... ... \n", "5648879 متى تجري الإنتخابات التشريعية في نيوزلندا ؟ \n", "5648880 هل يمكن علاج النمش عن طريق الجراحة ؟ \n", "5648881 ما هو دعاء الاستخارة ؟ \n", "5648882 كيف احضر عجينة البيتزا بالزبادي ؟ \n", "5648883 ما هي المساحة التي تشكلها قارة آسيا على الأرض ؟ \n", "\n", " positive \\\n", "0 المسألة الشرقية ( بالإنجليزية : Eastern Questi... \n", "1 بيت لحم ( بالسريانية : ؛ باليونانية : ؛ باللات... \n", "2 الإخوان المسلمون في سوريا : أسسها طلاب في ثلاث... \n", "3 كلينتون ريتشارد دوكينز عالم سلوك حيوان ، و عال... \n", "4 ضريبة الدخل هي ضريبة مباشرة تفرض على الأشخاص (... \n", "... ... \n", "5648879 متى تقوم الإنتخابات البرلمانية النيوزلندية ؟ \n", "5648880 كيف أتخلص من النمش من خلال الجراحة ؟ \n", "5648881 ما نص دعاء الاستخارة ؟ \n", "5648882 من طرق تحضير عجينة البيتزا بالزبادي ؟ \n", "5648883 ما المساحة التي تشكلها قارة آسيا على الأرض ؟ \n", "\n", " negative \n", "0 وفي 14 نوفمبر 1990 م ، وقعت الحكومة الألمانية ... \n", "1 كنائس بيت لحم تعتبر مدينة بيت لحم في فلسطين ، ... \n", "2 ذكر يوسف ندا ، مفوض العلاقات السياسية الدولية ... \n", "3 عام 2006 ، أسس دوكينز \" مؤسسة ريتشارد دوكينز ل... \n", "4 الضريبة على الشركات هي ضريبة مباشرة تفرضها الس... \n", "... ... \n", "5648879 متى وقعت غزوة تبوك ؟ \n", "5648880 ما هو علاج الصفار ؟ \n", "5648881 ما هو المشروع ؟ \n", "5648882 من طرق تحضير البيتزا الإيطالية ؟ \n", "5648883 ما العوامل التي تؤثر على التفويض ؟ \n", "\n", "[5648884 rows x 3 columns]" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "ds20" ] }, { "cell_type": "code", "execution_count": 8, "id": "8e8ecf55-c839-4cde-b0f5-f148250f8a95", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Done! Train: 3954218, Val: 1129777, Test: 564889\n" ] } ], "source": [ "\n", "# 1. Shuffle the entire dataset\n", "# frac=1 means 100% of the data, random_state ensures you can reproduce this later\n", "ds20_shuffled = ds20.sample(frac=1, random_state=42).reset_index(drop=True)\n", "\n", "# 2. Define your split points\n", "train_end = int(0.70 * len(ds20_shuffled))\n", "val_end = int(0.90 * len(ds20_shuffled)) # 70% + 20% = 90%\n", "\n", "# 3. Split the dataframe\n", "train_df = ds20_shuffled.iloc[:train_end]\n", "val_df = ds20_shuffled.iloc[train_end:val_end]\n", "test_df = ds20_shuffled.iloc[val_end:]\n", "\n", "# 4. Save to CSV\n", "# Using index=False so you don't get an extra 'unnamed' column later\n", "train_df.to_csv('train.csv', index=False)\n", "val_df.to_csv('val.csv', index=False)\n", "test_df.to_csv('test.csv', index=False)\n", "\n", "print(f\"Done! Train: {len(train_df)}, Val: {len(val_df)}, Test: {len(test_df)}\")" ] }, { "cell_type": "code", "execution_count": 9, "id": "e542d8ec-13d2-4b3d-be09-689b83f7dd41", "metadata": {}, "outputs": [], "source": [ "def showDataLayout():\n", " \"\"\"\n", " This function visualizes the distribution of the data across the training, validation, and test sets.\n", " It creates a bar chart showing the number of samples in each set.\n", " \"\"\"\n", " # Create a bar chart with the number of samples in each set\n", " plt.bar(\n", " [\"Train\", \"Valid\", \"Test\"], # Labels for the x-axis\n", " [len(train_df), len(val_df), len(test_df)], # Heights of the bars,\n", " #corresponding to the number of samples in each set\n", " align='center', # Align bars to the center\n", " color=['#3b528b', '#18b880', '#e6d74f'] # Colors for the bars\n", " )\n", " plt.legend() # Add a legend to the chart\n", "\n", " plt.ylabel('Number of images') # Label for the y-axis\n", " plt.title('Data distribution') # Title of the chart\n", "\n", " plt.show() # Display the chart\n", "\n" ] }, { "cell_type": "code", "execution_count": 10, "id": "0b375b75-fb06-43df-a551-aa4b7988386a", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/var/tmp/pbs.10156.head1/ipykernel_3650350/2531084511.py:14: UserWarning: No artists with labels found to put in legend. Note that artists whose label start with an underscore are ignored when legend() is called with no argument.\n", " plt.legend() # Add a legend to the chart\n" ] }, { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Call the function to display the data layout\n", "showDataLayout()" ] }, { "cell_type": "markdown", "id": "b30d510f-9d09-4a29-85a1-4b8753ab8196", "metadata": {}, "source": [ "# the code below was converted into a python file dont run it " ] }, { "cell_type": "code", "execution_count": 11, "id": "ab6e1c50-956e-45a1-a229-b48ab5b154e1", "metadata": {}, "outputs": [], "source": [ "train_batch_size = 128\n" ] }, { "cell_type": "code", "execution_count": 12, "id": "1aa3f4c8-f358-4c95-bf4c-b6b5680ec5ca", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2026-02-23 14:42:20 - Load pretrained SentenceTransformer: bert-base-arabertv02\n", "'[Errno -2] Name or service not known' thrown while requesting HEAD https://huggingface.co/bert-base-arabertv02/resolve/main/./modules.json\n", "2026-02-23 14:42:30 - '[Errno -2] Name or service not known' thrown while requesting HEAD https://huggingface.co/bert-base-arabertv02/resolve/main/./modules.json\n", "Retrying in 1s [Retry 1/5].\n", "2026-02-23 14:42:30 - Retrying in 1s [Retry 1/5].\n", "2026-02-23 14:42:31 - No sentence-transformers model found with name bert-base-arabertv02. Creating a new one with mean pooling.\n", "Loading weights: 100%|██████████| 199/199 [00:00<00:00, 1264.90it/s, Materializing param=pooler.dense.weight] \n", "\u001b[1mBertModel LOAD REPORT\u001b[0m from: bert-base-arabertv02\n", "Key | Status | | \n", "-------------------------------------------+------------+--+-\n", "cls.predictions.transform.dense.weight | UNEXPECTED | | \n", "cls.predictions.bias | UNEXPECTED | | \n", "cls.predictions.transform.LayerNorm.weight | UNEXPECTED | | \n", "cls.predictions.transform.dense.bias | UNEXPECTED | | \n", "cls.predictions.transform.LayerNorm.bias | UNEXPECTED | | \n", "cls.seq_relationship.weight | UNEXPECTED | | \n", "bert.embeddings.position_ids | UNEXPECTED | | \n", "cls.seq_relationship.bias | UNEXPECTED | | \n", "\n", "\u001b[3mNotes:\n", "- UNEXPECTED\u001b[3m\t:can be ignored when loading from different task/architecture; not ok if you expect identical arch.\u001b[0m\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Model is running on: cuda\n" ] } ], "source": [ "# Set the log level to INFO to get more information\n", "logging.basicConfig(format=\"%(asctime)s - %(message)s\", datefmt=\"%Y-%m-%d %H:%M:%S\", level=logging.INFO)\n", "\n", "# You can specify any Hugging Face pre-trained model here, for example, bert-base-uncased, roberta-base, xlm-roberta-base\n", "model_name = \"bert-base-arabertv02\"\n", "\n", "# Use a short nickname for the model to save space\n", "model_nickname = \"arabert\" \n", "timestamp = datetime.now().strftime(\"%Y%m%d_%H%M\") # Removed seconds to save space\n", "\n", "# Result: output/arabert_20260219_1711\n", "output_dir = f\"output/{model_nickname}_{timestamp}\"\n", "# Check if GPU is available, otherwise use CPU\n", "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n", "\n", "# 1. Here we define our SentenceTransformer model. If not already a Sentence Transformer model, it will automatically\n", "# create one with \"mean\" pooling.\n", "model = SentenceTransformer(model_name, device=device)\n", "print(f\"Model is running on: {device}\")" ] }, { "cell_type": "code", "execution_count": 13, "id": "239c132f-1932-47b8-a42a-a1ff35c99146", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2026-02-23 14:42:37 - Reading the training and eval dataset\n", "Generating train split: 3954218 examples [00:45, 87647.18 examples/s]\n", "Generating train split: 1129777 examples [00:12, 87451.65 examples/s]\n", "Generating train split: 564889 examples [00:06, 86963.25 examples/s]\n", "2026-02-23 14:43:59 - DatasetDict({\n", " train: Dataset({\n", " features: ['anchor', 'positive', 'negative'],\n", " num_rows: 3954218\n", " })\n", "})\n" ] } ], "source": [ "logging.info(\"Reading the training and eval dataset\")\n", "train_dataset = load_dataset(\"csv\", data_files=\"train.csv\")\n", "eval_dataset = load_dataset(\"csv\", data_files=\"val.csv\")\n", "test_dataset = load_dataset(\"csv\", data_files=\"test.csv\")\n", "logging.info(train_dataset)" ] }, { "cell_type": "code", "execution_count": 16, "id": "161c3598-9607-4037-81bb-f37d0578a5f2", "metadata": {}, "outputs": [], "source": [ "# 3. Define our training loss: \n", "matryoshka_dims = [768, 512, 256, 128, 64]\n", "inner_train_loss = losses.MultipleNegativesRankingLoss(model=model)\n", "train_loss = losses.MatryoshkaLoss(model, inner_train_loss, matryoshka_dims=matryoshka_dims)" ] }, { "cell_type": "code", "execution_count": 15, "id": "b369827e-015a-4129-886d-3caa0809c435", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2026-02-23 14:43:59 - TripletEvaluator: Evaluating the model on the dev-768 dataset (truncated to 768):\n" ] }, { "ename": "KeyboardInterrupt", "evalue": "", "output_type": "error", "traceback": [ "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", "\u001b[0;31mKeyboardInterrupt\u001b[0m Traceback (most recent call last)", "Cell \u001b[0;32mIn[15], line 14\u001b[0m\n\u001b[1;32m 12\u001b[0m dev_evaluator \u001b[38;5;241m=\u001b[39m SequentialEvaluator(evaluators, main_score_function\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mlambda\u001b[39;00m scores: scores[\u001b[38;5;241m0\u001b[39m])\n\u001b[1;32m 13\u001b[0m \u001b[38;5;66;03m#optional to comment in training \u001b[39;00m\n\u001b[0;32m---> 14\u001b[0m \u001b[43mdev_evaluator\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m)\u001b[49m\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/sentence_transformers/evaluation/SequentialEvaluator.py:44\u001b[0m, in \u001b[0;36mSequentialEvaluator.__call__\u001b[0;34m(self, model, output_path, epoch, steps)\u001b[0m\n\u001b[1;32m 42\u001b[0m scores \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 43\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m evaluator_idx, evaluator \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28menumerate\u001b[39m(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mevaluators):\n\u001b[0;32m---> 44\u001b[0m evaluation \u001b[38;5;241m=\u001b[39m \u001b[43mevaluator\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43moutput_path\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mepoch\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43msteps\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 46\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(evaluation, \u001b[38;5;28mdict\u001b[39m):\n\u001b[1;32m 47\u001b[0m scores\u001b[38;5;241m.\u001b[39mappend(evaluation)\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/sentence_transformers/evaluation/TripletEvaluator.py:179\u001b[0m, in \u001b[0;36mTripletEvaluator.__call__\u001b[0;34m(self, model, output_path, epoch, steps)\u001b[0m\n\u001b[1;32m 175\u001b[0m out_txt \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m (truncated to \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtruncate_dim\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m)\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m 177\u001b[0m logger\u001b[38;5;241m.\u001b[39minfo(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mTripletEvaluator: Evaluating the model on the \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mname\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m dataset\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mout_txt\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m:\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m--> 179\u001b[0m embeddings_anchors \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43membed_inputs\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43manchors\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 180\u001b[0m embeddings_positives \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39membed_inputs(model, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mpositives)\n\u001b[1;32m 181\u001b[0m embeddings_negatives \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39membed_inputs(model, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnegatives)\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/sentence_transformers/evaluation/TripletEvaluator.py:256\u001b[0m, in \u001b[0;36mTripletEvaluator.embed_inputs\u001b[0;34m(self, model, sentences, **kwargs)\u001b[0m\n\u001b[1;32m 250\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21membed_inputs\u001b[39m(\n\u001b[1;32m 251\u001b[0m \u001b[38;5;28mself\u001b[39m,\n\u001b[1;32m 252\u001b[0m model: SentenceTransformer,\n\u001b[1;32m 253\u001b[0m sentences: \u001b[38;5;28mstr\u001b[39m \u001b[38;5;241m|\u001b[39m \u001b[38;5;28mlist\u001b[39m[\u001b[38;5;28mstr\u001b[39m] \u001b[38;5;241m|\u001b[39m np\u001b[38;5;241m.\u001b[39mndarray,\n\u001b[1;32m 254\u001b[0m \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs,\n\u001b[1;32m 255\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m np\u001b[38;5;241m.\u001b[39mndarray:\n\u001b[0;32m--> 256\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mmodel\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mencode\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 257\u001b[0m \u001b[43m \u001b[49m\u001b[43msentences\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 258\u001b[0m \u001b[43m \u001b[49m\u001b[43mbatch_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbatch_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 259\u001b[0m \u001b[43m \u001b[49m\u001b[43mshow_progress_bar\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mshow_progress_bar\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 260\u001b[0m \u001b[43m \u001b[49m\u001b[43mconvert_to_numpy\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[1;32m 261\u001b[0m \u001b[43m \u001b[49m\u001b[43mtruncate_dim\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtruncate_dim\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 262\u001b[0m \u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 263\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/torch/utils/_contextlib.py:116\u001b[0m, in \u001b[0;36mcontext_decorator..decorate_context\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 113\u001b[0m \u001b[38;5;129m@functools\u001b[39m\u001b[38;5;241m.\u001b[39mwraps(func)\n\u001b[1;32m 114\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21mdecorate_context\u001b[39m(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs):\n\u001b[1;32m 115\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m ctx_factory():\n\u001b[0;32m--> 116\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/sentence_transformers/SentenceTransformer.py:1058\u001b[0m, in \u001b[0;36mSentenceTransformer.encode\u001b[0;34m(self, sentences, prompt_name, prompt, batch_size, show_progress_bar, output_value, precision, convert_to_numpy, convert_to_tensor, device, normalize_embeddings, truncate_dim, pool, chunk_size, **kwargs)\u001b[0m\n\u001b[1;32m 1056\u001b[0m all_embeddings \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 1057\u001b[0m length_sorted_idx \u001b[38;5;241m=\u001b[39m np\u001b[38;5;241m.\u001b[39margsort([\u001b[38;5;241m-\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_text_length(sen) \u001b[38;5;28;01mfor\u001b[39;00m sen \u001b[38;5;129;01min\u001b[39;00m sentences])\n\u001b[0;32m-> 1058\u001b[0m sentences_sorted \u001b[38;5;241m=\u001b[39m [sentences[\u001b[38;5;28mint\u001b[39m(idx)] \u001b[38;5;28;01mfor\u001b[39;00m idx \u001b[38;5;129;01min\u001b[39;00m length_sorted_idx]\n\u001b[1;32m 1060\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m start_index \u001b[38;5;129;01min\u001b[39;00m trange(\u001b[38;5;241m0\u001b[39m, \u001b[38;5;28mlen\u001b[39m(sentences), batch_size, desc\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mBatches\u001b[39m\u001b[38;5;124m\"\u001b[39m, disable\u001b[38;5;241m=\u001b[39m\u001b[38;5;129;01mnot\u001b[39;00m show_progress_bar):\n\u001b[1;32m 1061\u001b[0m sentences_batch \u001b[38;5;241m=\u001b[39m sentences_sorted[start_index : start_index \u001b[38;5;241m+\u001b[39m batch_size]\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/sentence_transformers/SentenceTransformer.py:1058\u001b[0m, in \u001b[0;36m\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m 1056\u001b[0m all_embeddings \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 1057\u001b[0m length_sorted_idx \u001b[38;5;241m=\u001b[39m np\u001b[38;5;241m.\u001b[39margsort([\u001b[38;5;241m-\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_text_length(sen) \u001b[38;5;28;01mfor\u001b[39;00m sen \u001b[38;5;129;01min\u001b[39;00m sentences])\n\u001b[0;32m-> 1058\u001b[0m sentences_sorted \u001b[38;5;241m=\u001b[39m [\u001b[43msentences\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;28;43mint\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43midx\u001b[49m\u001b[43m)\u001b[49m\u001b[43m]\u001b[49m \u001b[38;5;28;01mfor\u001b[39;00m idx \u001b[38;5;129;01min\u001b[39;00m length_sorted_idx]\n\u001b[1;32m 1060\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m start_index \u001b[38;5;129;01min\u001b[39;00m trange(\u001b[38;5;241m0\u001b[39m, \u001b[38;5;28mlen\u001b[39m(sentences), batch_size, desc\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mBatches\u001b[39m\u001b[38;5;124m\"\u001b[39m, disable\u001b[38;5;241m=\u001b[39m\u001b[38;5;129;01mnot\u001b[39;00m show_progress_bar):\n\u001b[1;32m 1061\u001b[0m sentences_batch \u001b[38;5;241m=\u001b[39m sentences_sorted[start_index : start_index \u001b[38;5;241m+\u001b[39m batch_size]\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/arrow_dataset.py:681\u001b[0m, in \u001b[0;36mColumn.__getitem__\u001b[0;34m(self, key)\u001b[0m\n\u001b[1;32m 679\u001b[0m source \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39msource\n\u001b[1;32m 680\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 681\u001b[0m source \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msource\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_fast_select_column\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcolumn_name\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 682\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m source[key][\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcolumn_name]\n\u001b[1;32m 683\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(key, \u001b[38;5;28mint\u001b[39m):\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/arrow_dataset.py:562\u001b[0m, in \u001b[0;36mtransmit_format..wrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 555\u001b[0m self_format \u001b[38;5;241m=\u001b[39m {\n\u001b[1;32m 556\u001b[0m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtype\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_format_type,\n\u001b[1;32m 557\u001b[0m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mformat_kwargs\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_format_kwargs,\n\u001b[1;32m 558\u001b[0m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcolumns\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_format_columns,\n\u001b[1;32m 559\u001b[0m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124moutput_all_columns\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_output_all_columns,\n\u001b[1;32m 560\u001b[0m }\n\u001b[1;32m 561\u001b[0m \u001b[38;5;66;03m# apply actual function\u001b[39;00m\n\u001b[0;32m--> 562\u001b[0m out: Union[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mDataset\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mDatasetDict\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 563\u001b[0m datasets: \u001b[38;5;28mlist\u001b[39m[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mDataset\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlist\u001b[39m(out\u001b[38;5;241m.\u001b[39mvalues()) \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(out, \u001b[38;5;28mdict\u001b[39m) \u001b[38;5;28;01melse\u001b[39;00m [out]\n\u001b[1;32m 564\u001b[0m \u001b[38;5;66;03m# re-apply format to the output\u001b[39;00m\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/arrow_dataset.py:2449\u001b[0m, in \u001b[0;36mDataset._fast_select_column\u001b[0;34m(self, column_name)\u001b[0m\n\u001b[1;32m 2446\u001b[0m \u001b[38;5;129m@transmit_format\u001b[39m\n\u001b[1;32m 2447\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m_fast_select_column\u001b[39m(\u001b[38;5;28mself\u001b[39m, column_name: \u001b[38;5;28mstr\u001b[39m) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mDataset\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n\u001b[1;32m 2448\u001b[0m dataset \u001b[38;5;241m=\u001b[39m copy\u001b[38;5;241m.\u001b[39mcopy(\u001b[38;5;28mself\u001b[39m)\n\u001b[0;32m-> 2449\u001b[0m dataset\u001b[38;5;241m.\u001b[39m_data \u001b[38;5;241m=\u001b[39m \u001b[43mdataset\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_data\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mselect\u001b[49m\u001b[43m(\u001b[49m\u001b[43m[\u001b[49m\u001b[43mcolumn_name\u001b[49m\u001b[43m]\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2450\u001b[0m dataset\u001b[38;5;241m.\u001b[39m_info \u001b[38;5;241m=\u001b[39m DatasetInfo(features\u001b[38;5;241m=\u001b[39mFeatures({column_name: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_info\u001b[38;5;241m.\u001b[39mfeatures[column_name]}))\n\u001b[1;32m 2451\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m dataset\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:1742\u001b[0m, in \u001b[0;36mConcatenationTable.select\u001b[0;34m(self, columns, *args, **kwargs)\u001b[0m\n\u001b[1;32m 1740\u001b[0m blocks \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 1741\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m tables \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mblocks:\n\u001b[0;32m-> 1742\u001b[0m blocks\u001b[38;5;241m.\u001b[39mappend([t\u001b[38;5;241m.\u001b[39mselect([c \u001b[38;5;28;01mfor\u001b[39;00m c \u001b[38;5;129;01min\u001b[39;00m columns \u001b[38;5;28;01mif\u001b[39;00m c \u001b[38;5;129;01min\u001b[39;00m t\u001b[38;5;241m.\u001b[39mcolumn_names], \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs) \u001b[38;5;28;01mfor\u001b[39;00m t \u001b[38;5;129;01min\u001b[39;00m tables])\n\u001b[1;32m 1743\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m ConcatenationTable(table, blocks)\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:1742\u001b[0m, in \u001b[0;36m\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m 1740\u001b[0m blocks \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 1741\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m tables \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mblocks:\n\u001b[0;32m-> 1742\u001b[0m blocks\u001b[38;5;241m.\u001b[39mappend([\u001b[43mt\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mselect\u001b[49m\u001b[43m(\u001b[49m\u001b[43m[\u001b[49m\u001b[43mc\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mfor\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mc\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;129;43;01min\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mcolumns\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mif\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mc\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;129;43;01min\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mt\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcolumn_names\u001b[49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m \u001b[38;5;28;01mfor\u001b[39;00m t \u001b[38;5;129;01min\u001b[39;00m tables])\n\u001b[1;32m 1743\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m ConcatenationTable(table, blocks)\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:1263\u001b[0m, in \u001b[0;36mMemoryMappedTable.select\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m 1261\u001b[0m replay \u001b[38;5;241m=\u001b[39m (\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mselect\u001b[39m\u001b[38;5;124m\"\u001b[39m, copy\u001b[38;5;241m.\u001b[39mdeepcopy(args), copy\u001b[38;5;241m.\u001b[39mdeepcopy(kwargs))\n\u001b[1;32m 1262\u001b[0m replays \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_append_replay(replay)\n\u001b[0;32m-> 1263\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mMemoryMappedTable\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtable\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mselect\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mpath\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mreplays\u001b[49m\u001b[43m)\u001b[49m\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:1011\u001b[0m, in \u001b[0;36mMemoryMappedTable.__init__\u001b[0;34m(self, table, path, replays)\u001b[0m\n\u001b[1;32m 1010\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m__init__\u001b[39m(\u001b[38;5;28mself\u001b[39m, table: pa\u001b[38;5;241m.\u001b[39mTable, path: \u001b[38;5;28mstr\u001b[39m, replays: Optional[\u001b[38;5;28mlist\u001b[39m[Replay]] \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m):\n\u001b[0;32m-> 1011\u001b[0m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__init__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mtable\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 1012\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mpath \u001b[38;5;241m=\u001b[39m os\u001b[38;5;241m.\u001b[39mpath\u001b[38;5;241m.\u001b[39mabspath(path)\n\u001b[1;32m 1013\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mreplays: \u001b[38;5;28mlist\u001b[39m[Replay] \u001b[38;5;241m=\u001b[39m replays \u001b[38;5;28;01mif\u001b[39;00m replays \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;28;01melse\u001b[39;00m []\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:166\u001b[0m, in \u001b[0;36mTable.__init__\u001b[0;34m(self, table)\u001b[0m\n\u001b[1;32m 165\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m__init__\u001b[39m(\u001b[38;5;28mself\u001b[39m, table: pa\u001b[38;5;241m.\u001b[39mTable):\n\u001b[0;32m--> 166\u001b[0m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__init__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mtable\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 167\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtable \u001b[38;5;241m=\u001b[39m table\n", "File \u001b[0;32m~/.conda/envs/sbert_khalil/lib/python3.10/site-packages/datasets/table.py:110\u001b[0m, in \u001b[0;36mIndexedTableMixin.__init__\u001b[0;34m(self, table)\u001b[0m\n\u001b[1;32m 106\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_schema: pa\u001b[38;5;241m.\u001b[39mSchema \u001b[38;5;241m=\u001b[39m table\u001b[38;5;241m.\u001b[39mschema\n\u001b[1;32m 107\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_batches: \u001b[38;5;28mlist\u001b[39m[pa\u001b[38;5;241m.\u001b[39mRecordBatch] \u001b[38;5;241m=\u001b[39m [\n\u001b[1;32m 108\u001b[0m recordbatch \u001b[38;5;28;01mfor\u001b[39;00m recordbatch \u001b[38;5;129;01min\u001b[39;00m table\u001b[38;5;241m.\u001b[39mto_batches() \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mlen\u001b[39m(recordbatch) \u001b[38;5;241m>\u001b[39m \u001b[38;5;241m0\u001b[39m\n\u001b[1;32m 109\u001b[0m ]\n\u001b[0;32m--> 110\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_offsets: np\u001b[38;5;241m.\u001b[39mndarray \u001b[38;5;241m=\u001b[39m \u001b[43mnp\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcumsum\u001b[49m\u001b[43m(\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;241;43m0\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m+\u001b[39;49m\u001b[43m \u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;28;43mlen\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mb\u001b[49m\u001b[43m)\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mfor\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mb\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;129;43;01min\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_batches\u001b[49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdtype\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mnp\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mint64\u001b[49m\u001b[43m)\u001b[49m\n", "\u001b[0;31mKeyboardInterrupt\u001b[0m: " ] } ], "source": [ "evaluators = []\n", "for dim in matryoshka_dims:\n", " evaluators.append(\n", " TripletEvaluator(\n", " anchors=eval_dataset[\"train\"][\"anchor\"],\n", " positives=eval_dataset[\"train\"][\"positive\"],\n", " negatives=eval_dataset[\"train\"][\"negative\"],\n", " name=f\"dev-{dim}\",\n", " truncate_dim=dim,\n", ")\n", " )\n", "dev_evaluator = SequentialEvaluator(evaluators, main_score_function=lambda scores: scores[0])\n", "#optional to comment in training \n", "dev_evaluator(model)" ] }, { "cell_type": "code", "execution_count": 17, "id": "6f1d4241-97be-4283-af24-cb6dd683b1b6", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "warmup_ratio is deprecated and will be removed in v5.2. Use `warmup_steps` instead.\n", "`logging_dir` is deprecated and will be removed in v5.2. Please set `TENSORBOARD_LOGGING_DIR` instead.\n" ] } ], "source": [ "args = SentenceTransformerTrainingArguments(\n", " # Required parameter:\n", " output_dir=output_dir,\n", " # Optional training parameters:\n", " num_train_epochs=1,\n", " per_device_train_batch_size=train_batch_size,\n", " per_device_eval_batch_size=train_batch_size,\n", " warmup_ratio=0.1,\n", " fp16=True, # Set to False if you get an error that your GPU can't run on FP16\n", " bf16=False, # Set to True if you have a GPU that supports BF16\n", " # Optional tracking/debugging parameters:\n", " batch_sampler=BatchSamplers.NO_DUPLICATES,\n", " report_to=\"tensorboard\", # Tell the trainer to log to TensorBoard\n", " logging_steps=5, # How often to log (every 10 steps)\n", " logging_dir=f\"{output_dir}/runs\", # Where to save the logs\n", " eval_strategy=\"steps\",\n", " eval_steps=10,\n", " save_strategy=\"steps\",\n", " save_steps=10,\n", " save_total_limit=2,\n", " \n", "\n", ")" ] }, { "cell_type": "code", "execution_count": 18, "id": "3e346f96-9379-416a-b95f-8d6084ecbb48", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ " " ] }, { "data": { "text/html": [ "\n", "
\n", " \n", " \n", " [ 11/247139 00:00 < 6:52:00, 10.00 it/s, Epoch 0.00/1]\n", "
\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
StepTraining LossValidation Loss

\n", "

\n", " \n", " \n", " [14319/70612 06:50 < 26:55, 34.84 it/s]\n", "
\n", " " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stderr", "output_type": "stream", "text": [ "\n", "KeyboardInterrupt\n", "\n" ] } ], "source": [ "# 6. Create the trainer & start training\n", "trainer = SentenceTransformerTrainer(\n", " model=model,\n", " args=args,\n", " train_dataset=train_dataset,\n", " eval_dataset=eval_dataset,\n", " loss=train_loss,\n", " evaluator=dev_evaluator,\n", ")\n", "trainer.train()" ] }, { "cell_type": "code", "execution_count": 30, "id": "b690f97f-0d80-4d53-9497-81c47a32913d", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'test-768_cosine_accuracy': 0.9560439586639404,\n", " 'test-512_cosine_accuracy': 0.9450549483299255,\n", " 'test-256_cosine_accuracy': 0.9450549483299255,\n", " 'test-128_cosine_accuracy': 0.9340659379959106,\n", " 'test-64_cosine_accuracy': 0.9230769276618958,\n", " 'sequential_score': 0.9230769276618958}" ] }, "execution_count": 30, "metadata": {}, "output_type": "execute_result" } ], "source": [ "evaluators = []\n", "for dim in matryoshka_dims:\n", " evaluators.append(\n", " TripletEvaluator(\n", " anchors=test_dataset[\"train\"][\"anchor\"],\n", " positives=test_dataset[\"train\"][\"positive\"],\n", " negatives=test_dataset[\"train\"][\"negative\"],\n", " name=f\"test-{dim}\",\n", " truncate_dim=dim,\n", ")\n", " )\n", "test_evaluator = SequentialEvaluator(evaluators)\n", "test_evaluator(model)" ] }, { "cell_type": "code", "execution_count": 31, "id": "73d7d7c8-dc94-435f-89e3-1b21c2c565e9", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 1.64it/s]\n" ] } ], "source": [ "# 8. Save the trained & evaluated model locally\n", "final_output_dir = f\"{output_dir}/final\"\n", "model.save(final_output_dir)\n" ] }, { "cell_type": "markdown", "id": "5756819d-f68d-4dea-85fe-5288321dc299", "metadata": {}, "source": [ "# preprocessing MS Marco dataset " ] }, { "cell_type": "code", "execution_count": 4, "id": "12f026af-e4cc-4c46-8a34-11c2c00428f2", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Shape before cleaning: (18606913, 3)\n", "Shape after cleaning: (18606913, 3)\n", "Preprocessing anchor...\n", "Preprocessing positive...\n", "Preprocessing negative...\n", "Done.\n" ] } ], "source": [ "# Load everything at once\n", "df = pd.read_csv(BASE / \"clean_dataset36.csv\", dtype=str, low_memory=False)\n", "\n", "# Rename\n", "df = df.rename(columns={\"query\": \"anchor\"})\n", "\n", "# Drop nulls and duplicates BEFORE preprocessing (saves time)\n", "print(f\"Shape before cleaning: {df.shape}\")\n", "df = df.dropna(subset=[\"anchor\", \"positive\", \"negative\"])\n", "df = df.drop_duplicates()\n", "print(f\"Shape after cleaning: {df.shape}\")\n", "\n", "# Preprocess\n", "for col in [\"anchor\", \"positive\", \"negative\"]:\n", " print(f\"Preprocessing {col}...\")\n", " df[col] = df[col].apply(lambda x: arabert_prep.preprocess(str(x)))\n", "\n", "# Shuffle\n", "df = df.sample(frac=1, random_state=42).reset_index(drop=True)\n", "\n", "# Split\n", "val_df = df.iloc[:10_000]\n", "train_df = df.iloc[10_000:]\n", "\n", "# Save\n", "val_df.to_csv(BASE / \"clean_dataset36_val.csv\", index=False)\n", "train_df.to_csv(BASE / \"clean_dataset36_train.csv\", index=False)\n", "\n", "# Delete original\n", "(BASE / \"clean_dataset36.csv\").unlink()\n", "\n", "print(\"Done.\")" ] }, { "cell_type": "code", "execution_count": 5, "id": "df6172b0-619d-41db-a3d3-ed4b84c6cf2b", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "RangeIndex: 10000 entries, 0 to 9999\n", "Data columns (total 3 columns):\n", " # Column Non-Null Count Dtype \n", "--- ------ -------------- ----- \n", " 0 anchor 10000 non-null object\n", " 1 positive 10000 non-null object\n", " 2 negative 10000 non-null object\n", "dtypes: object(3)\n", "memory usage: 234.5+ KB\n" ] } ], "source": [ "val_df.info()" ] }, { "cell_type": "code", "execution_count": 6, "id": "d701bc11-d2d7-45ed-bd4f-2749594e2712", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "RangeIndex: 18596913 entries, 10000 to 18606912\n", "Data columns (total 3 columns):\n", " # Column Dtype \n", "--- ------ ----- \n", " 0 anchor object\n", " 1 positive object\n", " 2 negative object\n", "dtypes: object(3)\n", "memory usage: 425.6+ MB\n" ] } ], "source": [ "train_df.info()" ] }, { "cell_type": "markdown", "id": "181a9dae-2b21-47b8-a75b-212197dc3998", "metadata": {}, "source": [ "# prepprocessing all datasets for multidataset training " ] }, { "cell_type": "code", "execution_count": 1, "id": "da8bed93-577f-4f69-b3a4-d977711896b3", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "WARNING:root:Model provided is not in the accepted model list. Preprocessor will default to a base Arabic preprocessor\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Found 8 CSV file(s) in /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data\n", "Loading AraBERT preprocessor...\n", "Ready.\n", "\n", "\n", "───────────────────────────────────────────────────────\n", " File : MultiNeg_30_ss.csv\n", " Rows (original) : 3,839\n", " Dropped (nulls) : 0 → 3,839 rows remain\n", " Dropped (dupes) : 0 → 3,839 rows remain\n", " Text columns : ['anchor', 'positive', 'negative_1', 'negative_2', 'negative_3', 'negative_4', 'negative_5']\n", " Skipped columns : []\n", " Preprocessing column: anchor ...\n", " Preprocessing column: positive ...\n", " Preprocessing column: negative_1 ...\n", " Preprocessing column: negative_2 ...\n", " Preprocessing column: negative_3 ...\n", " Preprocessing column: negative_4 ...\n", " Preprocessing column: negative_5 ...\n", " ✓ Saved (3,839 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/MultiNeg_30_ss.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : MultiNeg_4_ss.csv\n", " Rows (original) : 500\n", " Dropped (nulls) : 0 → 500 rows remain\n", " Dropped (dupes) : 0 → 500 rows remain\n", " Text columns : ['anchor', 'positive', 'negative_1', 'negative_2', 'negative_3', 'negative_4']\n", " Skipped columns : []\n", " Preprocessing column: anchor ...\n", " Preprocessing column: positive ...\n", " Preprocessing column: negative_1 ...\n", " Preprocessing column: negative_2 ...\n", " Preprocessing column: negative_3 ...\n", " Preprocessing column: negative_4 ...\n", " ✓ Saved (500 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/MultiNeg_4_ss.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : a_p_n_ss.csv\n", " Rows (original) : 3,538,680\n", " Dropped (nulls) : 0 → 3,538,680 rows remain\n", " Dropped (dupes) : 32,501 → 3,506,179 rows remain\n", " Text columns : ['anchor', 'positive', 'negative']\n", " Skipped columns : []\n", " Preprocessing column: anchor ...\n", " Preprocessing column: positive ...\n", " Preprocessing column: negative ...\n", " ✓ Saved (3,506,179 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/a_p_n_ss.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : a_p_n_sts.csv\n", " Rows (original) : 5,627,815\n", " Dropped (nulls) : 0 → 5,627,815 rows remain\n", " Dropped (dupes) : 1,171,039 → 4,456,776 rows remain\n", " Text columns : ['anchor', 'positive', 'negative']\n", " Skipped columns : []\n", " Preprocessing column: anchor ...\n", " Preprocessing column: positive ...\n", " Preprocessing column: negative ...\n", " ✓ Saved (4,456,776 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/a_p_n_sts.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : a_p_ss.csv\n", " Rows (original) : 277,638\n", " Dropped (nulls) : 0 → 277,638 rows remain\n", " Dropped (dupes) : 620 → 277,018 rows remain\n", " Text columns : ['anchor', 'positive']\n", " Skipped columns : []\n", " Preprocessing column: anchor ...\n", " Preprocessing column: positive ...\n", " ✓ Saved (277,018 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/a_p_ss.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : s1_s2_label_ss.csv\n", " Rows (original) : 436\n", " Dropped (nulls) : 0 → 436 rows remain\n", " Dropped (dupes) : 0 → 436 rows remain\n", " Text columns : ['sentence1', 'sentence2']\n", " Skipped columns : ['label']\n", " Preprocessing column: sentence1 ...\n", " Preprocessing column: sentence2 ...\n", " ✓ Saved (436 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/s1_s2_label_ss.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : s1_s2_label_sts.csv\n", " Rows (original) : 15,712\n", " Dropped (nulls) : 0 → 15,712 rows remain\n", " Dropped (dupes) : 0 → 15,712 rows remain\n", " Text columns : ['sentence1', 'sentence2']\n", " Skipped columns : ['label']\n", " Preprocessing column: sentence1 ...\n", " Preprocessing column: sentence2 ...\n", " ✓ Saved (15,712 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/s1_s2_label_sts.csv\n", "\n", "───────────────────────────────────────────────────────\n", " File : s1_s2_score_sts.csv\n", " Rows (original) : 7,982\n", " Dropped (nulls) : 0 → 7,982 rows remain\n", " Dropped (dupes) : 0 → 7,982 rows remain\n", " Text columns : ['sentence1', 'sentence2']\n", " Skipped columns : ['score']\n", " Preprocessing column: sentence1 ...\n", " Preprocessing column: sentence2 ...\n", " ✓ Saved (7,982 rows) → /home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data/s1_s2_score_sts.csv\n", "\n", "═══════════════════════════════════════════════════════\n", " SUMMARY\n", "═══════════════════════════════════════════════════════\n", " file original_rows dropped_nulls dropped_dupes final_rows text_columns skipped_columns\n", " MultiNeg_30_ss.csv 3839 0 0 3839 anchor, positive, negative_1, negative_2, negative_3, negative_4, negative_5 \n", " MultiNeg_4_ss.csv 500 0 0 500 anchor, positive, negative_1, negative_2, negative_3, negative_4 \n", " a_p_n_ss.csv 3538680 0 32501 3506179 anchor, positive, negative \n", " a_p_n_sts.csv 5627815 0 1171039 4456776 anchor, positive, negative \n", " a_p_ss.csv 277638 0 620 277018 anchor, positive \n", " s1_s2_label_ss.csv 436 0 0 436 sentence1, sentence2 label\n", "s1_s2_label_sts.csv 15712 0 0 15712 sentence1, sentence2 label\n", "s1_s2_score_sts.csv 7982 0 0 7982 sentence1, sentence2 score\n", "\n", " Total rows across all files : 9,472,602\n", " Total dropped : 1,204,160\n", " Total remaining : 8,268,442\n", "═══════════════════════════════════════════════════════\n" ] } ], "source": [ "\"\"\"\n", "Preprocessing Pipeline\n", "======================\n", "For every CSV in the target directory:\n", " 1. Drop rows with any null value\n", " 2. Drop duplicate rows (within each file)\n", " 3. Apply AraBERT preprocessing to all text columns (skips label/score columns)\n", " 4. Shuffle\n", " 5. Overwrite the original file\n", "\n", "Usage\n", "-----\n", "pip install arabert pandas\n", "python preprocess_clean_data.py\n", "\"\"\"\n", "\n", "import pandas as pd\n", "from pathlib import Path\n", "from arabert.preprocess import ArabertPreprocessor\n", "\n", "# ─────────────────────────────────────────────\n", "# CONFIG\n", "# ─────────────────────────────────────────────\n", "\n", "DATA_DIR = Path(\"/home/skiredj.abderrahman/khalil/sbert_training/third_training/clean_data\")\n", "MODEL_NAME = \"/home/skiredj.abderrahman/khalil/sbert_training/second_training/bert-base-arabertv02/\" # change if using a different AraBERT variant\n", "RANDOM_STATE = 42\n", "\n", "\n", "# ─────────────────────────────────────────────\n", "# HELPERS\n", "# ─────────────────────────────────────────────\n", "\n", "def is_text_column(series: pd.Series) -> bool:\n", " \"\"\"\n", " Returns True if the column contains real Arabic text.\n", " Skips columns that are numeric, binary labels (0/1),\n", " or float scores — even if stored as strings.\n", " \"\"\"\n", " # Pandas already knows it's numeric\n", " if pd.api.types.is_numeric_dtype(series):\n", " return False\n", "\n", " # Sample up to 50 non-null values and check if they all parse as numbers\n", " sample = series.dropna().astype(str).head(50)\n", " if len(sample) == 0:\n", " return False\n", "\n", " def looks_numeric(val: str) -> bool:\n", " try:\n", " float(val)\n", " return True\n", " except ValueError:\n", " return False\n", "\n", " if all(looks_numeric(v) for v in sample):\n", " return False\n", "\n", " return True\n", "\n", "\n", "def preprocess_file(path: Path, arabert_prep: ArabertPreprocessor) -> dict:\n", " \"\"\"\n", " Full pipeline for a single CSV file.\n", " Returns a summary dict for the final report.\n", " \"\"\"\n", " filename = path.name\n", " print(f\"\\n{'─'*55}\")\n", " print(f\" File : {filename}\")\n", "\n", " df = pd.read_csv(path, low_memory=False)\n", " original_rows = len(df)\n", " print(f\" Rows (original) : {original_rows:,}\")\n", "\n", " # ── 1. Drop nulls ────────────────────────\n", " df = df.dropna()\n", " after_null = len(df)\n", " dropped_null = original_rows - after_null\n", " print(f\" Dropped (nulls) : {dropped_null:,} → {after_null:,} rows remain\")\n", "\n", " # ── 2. Drop duplicates ───────────────────\n", " df = df.drop_duplicates()\n", " after_dedup = len(df)\n", " dropped_dup = after_null - after_dedup\n", " print(f\" Dropped (dupes) : {dropped_dup:,} → {after_dedup:,} rows remain\")\n", "\n", " # ── 3. Identify text columns ─────────────\n", " text_cols = [col for col in df.columns if is_text_column(df[col])]\n", " skip_cols = [col for col in df.columns if col not in text_cols]\n", " print(f\" Text columns : {text_cols}\")\n", " print(f\" Skipped columns : {skip_cols}\")\n", "\n", " # ── 4. AraBERT preprocessing ─────────────\n", " for col in text_cols:\n", " print(f\" Preprocessing column: {col} ...\")\n", " df[col] = df[col].apply(lambda x: arabert_prep.preprocess(str(x)))\n", "\n", " # ── 5. Shuffle ───────────────────────────\n", " df = df.sample(frac=1, random_state=RANDOM_STATE).reset_index(drop=True)\n", "\n", " # ── 6. Overwrite original ────────────────\n", " df.to_csv(path, index=False)\n", " print(f\" ✓ Saved ({after_dedup:,} rows) → {path}\")\n", "\n", " return {\n", " \"file\": filename,\n", " \"original_rows\": original_rows,\n", " \"dropped_nulls\": dropped_null,\n", " \"dropped_dupes\": dropped_dup,\n", " \"final_rows\": after_dedup,\n", " \"text_columns\": \", \".join(text_cols),\n", " \"skipped_columns\": \", \".join(skip_cols),\n", " }\n", "\n", "\n", "# ─────────────────────────────────────────────\n", "# MAIN\n", "# ─────────────────────────────────────────────\n", "\n", "def main():\n", " csv_files = sorted(DATA_DIR.glob(\"*.csv\"))\n", " if not csv_files:\n", " print(f\"No CSV files found in {DATA_DIR}\")\n", " return\n", "\n", " print(f\"Found {len(csv_files)} CSV file(s) in {DATA_DIR}\")\n", " print(\"Loading AraBERT preprocessor...\")\n", " arabert_prep = ArabertPreprocessor(model_name=MODEL_NAME)\n", " print(\"Ready.\\n\")\n", "\n", " summaries = []\n", " for path in csv_files:\n", " try:\n", " summary = preprocess_file(path, arabert_prep)\n", " summaries.append(summary)\n", " except Exception as e:\n", " print(f\" ✗ ERROR processing {path.name}: {e}\")\n", " summaries.append({\n", " \"file\": path.name,\n", " \"original_rows\": \"ERROR\",\n", " \"dropped_nulls\": \"ERROR\",\n", " \"dropped_dupes\": \"ERROR\",\n", " \"final_rows\": \"ERROR\",\n", " \"text_columns\": \"\",\n", " \"skipped_columns\": str(e),\n", " })\n", "\n", " # ── Print final summary ──────────────────\n", " print(f\"\\n{'═'*55}\")\n", " print(\" SUMMARY\")\n", " print(f\"{'═'*55}\")\n", " summary_df = pd.DataFrame(summaries)\n", " print(summary_df.to_string(index=False))\n", "\n", " total_original = sum(s[\"original_rows\"] for s in summaries if isinstance(s[\"original_rows\"], int))\n", " total_final = sum(s[\"final_rows\"] for s in summaries if isinstance(s[\"final_rows\"], int))\n", " total_dropped = total_original - total_final\n", " print(f\"\\n Total rows across all files : {total_original:,}\")\n", " print(f\" Total dropped : {total_dropped:,}\")\n", " print(f\" Total remaining : {total_final:,}\")\n", " print(f\"{'═'*55}\")\n", "\n", "\n", "if __name__ == \"__main__\":\n", " main()" ] }, { "cell_type": "code", "execution_count": null, "id": "ab0ddcc3-40d6-432b-808e-6aa366305684", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "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": "3.10.19" } }, "nbformat": 4, "nbformat_minor": 5 }