{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"This notebook shows how to fine-tune a BERT model (from huggingface) for our dataset recognition task.\n\nNote that internet is needed during the training phase (for downloading the bert-base-cased model). Internet can be turned off during prediction.","metadata":{}},{"cell_type":"markdown","source":"## Install packages","metadata":{}},{"cell_type":"code","source":"!pip install datasets --no-index --find-links=file:///kaggle/input/coleridge-packages/packages/datasets\n!pip install ../input/coleridge-packages/seqeval-1.2.2-py3-none-any.whl\n!pip install ../input/coleridge-packages/tokenizers-0.10.1-cp37-cp37m-manylinux1_x86_64.whl\n!pip install ../input/coleridge-packages/transformers-4.5.0.dev0-py3-none-any.whl","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2021-06-07T12:10:11.061948Z","iopub.execute_input":"2021-06-07T12:10:11.062343Z","iopub.status.idle":"2021-06-07T12:10:38.693167Z","shell.execute_reply.started":"2021-06-07T12:10:11.062309Z","shell.execute_reply":"2021-06-07T12:10:38.69231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Import","metadata":{}},{"cell_type":"code","source":"import os\nimport re\nimport json\nimport time\nimport datetime\nimport random\nimport glob\nimport importlib\n\nimport numpy as np\nimport pandas as pd\n\nfrom tqdm import tqdm\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nrandom.seed(123)\nnp.random.seed(456)","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:10:51.75088Z","iopub.execute_input":"2021-06-07T12:10:51.751215Z","iopub.status.idle":"2021-06-07T12:10:52.470621Z","shell.execute_reply.started":"2021-06-07T12:10:51.751184Z","shell.execute_reply":"2021-06-07T12:10:52.469869Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# copy my_seqeval.py to the working directory because the input directory is non-writable\n!cp /kaggle/input/coleridge-packages/my_seqeval.py ./","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:10:56.211313Z","iopub.execute_input":"2021-06-07T12:10:56.211636Z","iopub.status.idle":"2021-06-07T12:10:56.957769Z","shell.execute_reply.started":"2021-06-07T12:10:56.211608Z","shell.execute_reply":"2021-06-07T12:10:56.956689Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Hyper-parameters","metadata":{}},{"cell_type":"code","source":"MAX_LENGTH = 64 # max no. words for each sentence.\nOVERLAP = 20 # if a sentence exceeds MAX_LENGTH, we split it to multiple sentences with overlapping\n\nMAX_SAMPLE = None # set a small number for experimentation, set None for production.","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:11:28.253717Z","iopub.execute_input":"2021-06-07T12:11:28.254082Z","iopub.status.idle":"2021-06-07T12:11:28.258242Z","shell.execute_reply.started":"2021-06-07T12:11:28.254047Z","shell.execute_reply":"2021-06-07T12:11:28.257446Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Load data","metadata":{}},{"cell_type":"code","source":"COMPUTE_EVAL = True","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:15:56.970424Z","iopub.execute_input":"2021-06-07T12:15:56.970759Z","iopub.status.idle":"2021-06-07T12:15:56.976979Z","shell.execute_reply.started":"2021-06-07T12:15:56.970727Z","shell.execute_reply":"2021-06-07T12:15:56.976261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if COMPUTE_EVAL:\n    df = pd.read_csv('../input/coleridgeinitiative-show-us-the-data/train.csv')\n    np.random.seed(456)\n    df['split'] = np.random.randn(df.shape[0], 1)\n    \n    np.random.seed(456)\n    msk = np.random.rand(len(df)) <= 0.9\n\n    train = df[msk]\n    train2 = train\n    val = df[~msk]\nelse:\n    train_path = '../input/coleridgeinitiative-show-us-the-data/train.csv'\n    train = pd.read_csv(train_path)\n    train = train[:MAX_SAMPLE]\nprint(f'No. raw training rows: {len(train)}')","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:21:00.864612Z","iopub.execute_input":"2021-06-07T12:21:00.864964Z","iopub.status.idle":"2021-06-07T12:21:00.925276Z","shell.execute_reply.started":"2021-06-07T12:21:00.864934Z","shell.execute_reply":"2021-06-07T12:21:00.924369Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# print(\"Length oftraining data: \", str(len(train)))\n# print(train['Id'])\n# print(train['Id'][1])\n# print(train['Id'][2])\n# print(train['Id'][3])\n# print(train['Id'][4])\n# print(train['Id'][5])\n# print(train['Id'][6])","metadata":{"execution":{"iopub.status.busy":"2021-06-07T12:21:02.783451Z","iopub.execute_input":"2021-06-07T12:21:02.78378Z","iopub.status.idle":"2021-06-07T12:21:02.796216Z","shell.execute_reply.started":"2021-06-07T12:21:02.78375Z","shell.execute_reply":"2021-06-07T12:21:02.795176Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Group by publication, training labels should have the same form as expected output.","metadata":{}},{"cell_type":"code","source":"train = train.groupby('Id').agg({\n    'pub_title': 'first',\n    'dataset_title': '|'.join,\n    'dataset_label': '|'.join,\n    'cleaned_label': '|'.join\n}).reset_index()\n\nprint(f'No. grouped training rows: {len(train)}')","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:11:33.188002Z","iopub.execute_input":"2021-05-31T15:11:33.188345Z","iopub.status.idle":"2021-05-31T15:11:33.580003Z","shell.execute_reply.started":"2021-05-31T15:11:33.188313Z","shell.execute_reply":"2021-05-31T15:11:33.578287Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"paper_train_folder = '../input/coleridgeinitiative-show-us-the-data/train'\npapers = {}\nfor paper_id in train['Id'].unique():\n    with open(f'{paper_train_folder}/{paper_id}.json', 'r') as f:\n        paper = json.load(f)\n        papers[paper_id] = paper","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:11:37.665923Z","iopub.execute_input":"2021-05-31T15:11:37.66628Z","iopub.status.idle":"2021-05-31T15:12:35.574114Z","shell.execute_reply.started":"2021-05-31T15:11:37.666247Z","shell.execute_reply":"2021-05-31T15:12:35.57328Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Transform data to NER format","metadata":{}},{"cell_type":"code","source":"def clean_training_text(txt):\n    \"\"\"\n    similar to the default clean_text function but without lowercasing.\n    \"\"\"\n    return re.sub('[^A-Za-z0-9]+', ' ', str(txt)).strip()\n\ndef shorten_sentences(sentences):\n    short_sentences = []\n    for sentence in sentences:\n        words = sentence.split()\n        if len(words) > MAX_LENGTH:\n            for p in range(0, len(words), MAX_LENGTH - OVERLAP):\n                short_sentences.append(' '.join(words[p:p+MAX_LENGTH]))\n        else:\n            short_sentences.append(sentence)\n    return short_sentences\n\ndef find_sublist(big_list, small_list):\n    all_positions = []\n    for i in range(len(big_list) - len(small_list) + 1):\n        if small_list == big_list[i:i+len(small_list)]:\n            all_positions.append(i)\n    \n    return all_positions\n\ndef tag_sentence(sentence, labels): # requirement: both sentence and labels are already cleaned\n    sentence_words = sentence.split()\n    \n    if labels is not None and any(re.findall(f'\\\\b{label}\\\\b', sentence)\n                                  for label in labels): # positive sample\n        nes = ['O'] * len(sentence_words)\n        for label in labels:\n            label_words = label.split()\n\n            all_pos = find_sublist(sentence_words, label_words)\n            for pos in all_pos:\n                nes[pos] = 'B'\n                for i in range(pos+1, pos+len(label_words)):\n                    nes[i] = 'I'\n\n        return True, list(zip(sentence_words, nes))\n        \n    else: # negative sample\n        nes = ['O'] * len(sentence_words)\n        return False, list(zip(sentence_words, nes))","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:12:35.575614Z","iopub.execute_input":"2021-05-31T15:12:35.575962Z","iopub.status.idle":"2021-05-31T15:12:35.586689Z","shell.execute_reply.started":"2021-05-31T15:12:35.575925Z","shell.execute_reply":"2021-05-31T15:12:35.585859Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cnt_pos, cnt_neg = 0, 0 # number of sentences that contain/not contain labels\nner_data = []\n\npbar = tqdm(total=len(train))\nfor i, id, dataset_label in train[['Id', 'dataset_label']].itertuples():\n    # paper\n    paper = papers[id]\n    \n    # labels\n    labels = dataset_label.split('|')\n    labels = [clean_training_text(label) for label in labels]\n    \n    # sentences\n    sentences = set([clean_training_text(sentence) for section in paper \n                 for sentence in section['text'].split('.') \n                ])\n    sentences = shorten_sentences(sentences) # make sentences short\n    sentences = [sentence for sentence in sentences if len(sentence) > 10] # only accept sentences with length > 10 chars\n    \n    # positive sample\n    for sentence in sentences:\n        is_positive, tags = tag_sentence(sentence, labels)\n        if is_positive:\n            cnt_pos += 1\n            ner_data.append(tags)\n        elif any(word in sentence.lower() for word in ['data', 'study']): \n            ner_data.append(tags)\n            cnt_neg += 1\n    \n    # process bar\n    pbar.update(1)\n    pbar.set_description(f\"Training data size: {cnt_pos} positives + {cnt_neg} negatives\")\n\n# shuffling\nrandom.shuffle(ner_data)","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:12:55.054733Z","iopub.execute_input":"2021-05-31T15:12:55.055053Z","iopub.status.idle":"2021-05-31T15:15:35.826174Z","shell.execute_reply.started":"2021-05-31T15:12:55.055021Z","shell.execute_reply":"2021-05-31T15:15:35.825385Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"write data to file.","metadata":{}},{"cell_type":"code","source":"with open('train_ner.json', 'w') as f:\n    for row in ner_data:\n        words, nes = list(zip(*row))\n        row_json = {'tokens' : words, 'tags' : nes}\n        json.dump(row_json, f)\n        f.write('\\n')","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:19:08.155355Z","iopub.execute_input":"2021-05-31T15:19:08.155679Z","iopub.status.idle":"2021-05-31T15:19:37.751435Z","shell.execute_reply.started":"2021-05-31T15:19:08.15565Z","shell.execute_reply":"2021-05-31T15:19:37.750635Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Fine-tune a BERT model for NER","metadata":{}},{"cell_type":"code","source":"#Default:\n# !python ../input/kaggle-ner-utils/kaggle_run_ner.py \\\n# --model_name_or_path 'bert-base-cased' \\\n# --train_file './train_ner.json' \\\n# --validation_file './train_ner.json' \\\n# --num_train_epochs 1 \\\n# --per_device_train_batch_size 8 \\\n# --per_device_eval_batch_size 8 \\\n# --save_steps 15000 \\\n# --output_dir './output' \\\n# --report_to 'none' \\\n# --seed 123 \\\n# --do_train ","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#SciBERT:\n# !python ../input/kaggle-ner-utils/kaggle_run_ner.py \\\n# --model_name_or_path 'allenai/scibert_scivocab_uncased' \\\n# --train_file './train_ner.json' \\\n# --validation_file './train_ner.json' \\\n# --num_train_epochs 1 \\\n# --per_device_train_batch_size 8 \\\n# --per_device_eval_batch_size 8 \\\n# --save_steps 15000 \\\n# --output_dir './output' \\\n# --report_to 'none' \\\n# --seed 123 \\\n# --do_train ","metadata":{"execution":{"iopub.status.busy":"2021-05-31T15:20:01.201126Z","iopub.execute_input":"2021-05-31T15:20:01.2015Z","iopub.status.idle":"2021-05-31T19:19:29.063282Z","shell.execute_reply.started":"2021-05-31T15:20:01.20147Z","shell.execute_reply":"2021-05-31T19:19:29.062362Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#RoBERTa\n!python ../input/kaggle-ner-utils/kaggle_run_ner.py \\\n--model_name_or_path 'roberta-base' \\\n--train_file './train_ner.json' \\\n--validation_file './train_ner.json' \\\n--num_train_epochs 1 \\\n--per_device_train_batch_size 8 \\\n--per_device_eval_batch_size 8 \\\n--save_steps 15000 \\\n--output_dir './output' \\\n--report_to 'none' \\\n--seed 123 \\\n--do_train ","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"After the tuning finishes, we should find our model in './output'.","metadata":{}}]}