{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import json\nimport os\nimport heapq\n\nimport numpy as np\nfrom sklearn.feature_extraction.text import TfidfVectorizer\nfrom sklearn.metrics.pairwise import cosine_similarity","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"path_train = '/kaggle/input/tensorflow2-question-answering/simplified-nq-train.jsonl'\ntrain = []\nwith open(path_train, 'r') as file:\n    for i in range(1000):\n        train.append(json.loads(file.readline()))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train[0].keys()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train[0]['long_answer_candidates'][:5]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def extract_corpus(doc):\n    document_text = doc['document_text']\n    long_answer_candidates = doc['long_answer_candidates']\n    tokens = document_text.split(' ')\n    corpus = []\n    for candidate in long_answer_candidates:\n        start_token = candidate['start_token']\n        end_token = candidate['end_token']\n        corpus.append(\" \".join(tokens[start_token:end_token]))\n    return corpus\n\ndef get_long_answer(doc):\n    document_text = doc['document_text']\n    tokens = document_text.split(' ')\n    # even though annotatations is an array, it seems to be all length of 1\n    long_answer_anno = doc['annotations'][0]['long_answer']\n    start_token = long_answer_anno['start_token']\n    end_token = long_answer_anno['end_token']\n    long_answer = \" \".join(tokens[start_token:end_token])\n    return long_answer","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# testing out extract_corpus\ncorpus = extract_corpus(train[0])\nlen(corpus)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"corpus[0]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"corpus[1]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"corpus[2]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_top_n_candidates(corpus, question_text, n):\n    tfidf = TfidfVectorizer(stop_words='english')\n    X_corpus = tfidf.fit_transform(corpus)\n    X_question_text = tfidf.transform([question_text])\n    similarity = cosine_similarity(X_corpus, X_question_text)\n    top_n_idx = heapq.nlargest(n, range(len(similarity)), similarity.take)\n    top_n_candidates = [corpus[i] for i in top_n_idx]\n    return top_n_candidates","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def print_ranking(doc, n):\n    question_text = doc['question_text']\n    print('Question:')\n    print(question_text)\n    print()\n\n    long_answer = get_long_answer(doc)\n    print('Expected long answer:')\n    print(long_answer)\n    print()\n\n    corpus = extract_corpus(doc)\n    top_candidates = get_top_n_candidates(corpus, question_text, n)\n    print('Ranked long answers:')\n    found = False\n    for idx, candidate in enumerate(top_candidates):\n        if long_answer == candidate:\n            print(\"CORRECT LONG ANSWER FOUND :)\")\n            found = True\n        print(f\"#{idx + 1}:\")\n        print(candidate)\n        print()\n    if not found:\n        print(\"correct long answer not found :(\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"print_ranking(train[0], 3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"print_ranking(train[1], 3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"print_ranking(train[2], 3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"print_ranking(train[3], 3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def find_long_answer(doc, n):\n    question_text = doc['question_text']\n    long_answer = get_long_answer(doc)\n    corpus = extract_corpus(doc)\n    top_candidates = get_top_n_candidates(corpus, question_text, n)\n    candidate_match = [candidate for candidate in top_candidates if long_answer == candidate]\n    found = True if len(candidate_match) > 0 else False\n    return found\n    \ndef calc_find_score(docs, n):\n    num_found = 0\n    for doc in docs:\n        if find_long_answer(doc, n):\n            num_found += 1\n    return num_found / len(docs)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"calc_find_score(train, 3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"calc_find_score(train, 5)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"calc_find_score(train, 10)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"calc_find_score(train, 20)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"","execution_count":null,"outputs":[]}],"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":4,"nbformat_minor":1}