{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import os, gc\nfrom fastai.text import *\nfrom tqdm import tqdm_notebook as tqdm\nprint(os.listdir(\"../input\"))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"c48c17827846b5be89c471f5cd0de9d092899e54"},"cell_type":"code","source":"# make training deterministic/reproducible\ndef seed_everything(seed=2018):\n    random.seed(seed)\n    torch.manual_seed(seed)\n    torch.cuda.manual_seed_all(seed)\n    np.random.seed(seed)\n    os.environ['PYTHONHASHSEED'] = str(seed)\n    torch.backends.cudnn.deterministic = True\nseed_everything()\n\ndef f1_score(y_pred, targets):\n    epsilon = 1e-07\n    \n    y_pred = y_pred.argmax(dim=1)\n    targets = targets.argmax(dim=1)\n\n    tp = (y_pred*targets).float().sum(dim=0)\n    tn = ((1-targets)*(1-y_pred)).float().sum(dim=0)\n    fp = ((1-targets)*y_pred).float().sum(dim=0)\n    fn = (targets*(1-y_pred)).sum(dim=0)\n\n    p = tp / (tp + fp + epsilon)\n    r = tp / (tp + fn + epsilon)\n\n    f1 = 2*p*r / (p+r+epsilon)\n    f1 = torch.where(f1!=f1, torch.zeros_like(f1), f1)\n    return f1.mean()","execution_count":null,"outputs":[]},{"metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","trusted":true},"cell_type":"code","source":"EMBED_SIZE = 50\nMAX_FEATURES = 60000\nMAX_LENGTH = 100\nEMBEDDING_FILE = '../input/embeddings/wiki-news-300d-1M/wiki-news-300d-1M.vec'","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"19534579b5ebeb831b4391b06136b8fddc1e9c6e"},"cell_type":"code","source":"# df = pd.read_csv('../input/train.csv')\n\n# insincere_df = df[df.target==1]\n# sincere_df = df[df.target==0]\n\n# sincere_df = sincere_df.iloc[np.random.permutation(len(sincere_df))]\n# sincere_df = sincere_df[:int(len(insincere_df)*5)]\n\n# del df\n\n# df = pd.concat([insincere_df, sincere_df])\n# df = df.iloc[np.random.permutation(len(df))]\n\n# del insincere_df\n# del sincere_df\n# gc.collect()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"07589f4184b3547306afecb73e2177c04f0f4850"},"cell_type":"code","source":"train_df = pd.read_csv('../input/train.csv')\ntest_df = pd.read_csv('../input/test.csv')\ntrain_df.head()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"c3c51e03b5a24e4dac3667a4392af60f63e7271c"},"cell_type":"code","source":"def truncate(df):\n    df['question_text'] = df.question_text.apply(lambda x: x[:MAX_LENGTH])\n    \ntruncate(train_df)\ntruncate(test_df)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"d456198e89e5d40fd5f707dcad8eb5d5b313a6ae"},"cell_type":"code","source":"train_df = train_df.iloc[np.random.permutation(len(train_df))]\ncut = int(0.2 * len(train_df)) + 1\ntrain_df, valid_df = train_df[cut:], train_df[:cut]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"f348dc0717c08af292bd4f139e75dedb90585fd7"},"cell_type":"code","source":"%%time\ndata = TextDataBunch.from_df(path='.',\n                             train_df=train_df, \n                             valid_df=valid_df,\n                             test_df=test_df,\n                             text_cols='question_text', \n                             label_cols='target',\n                             max_vocab=MAX_FEATURES)\nprint(len(data.vocab.itos))\ndata.save()\ndel train_df\ndel valid_df \ndel test_df \ndel data\ngc.collect()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"d494ef801ff80e5c18de7c0ce4e7fd0af28bbb3b"},"cell_type":"code","source":"%%time\ndata = TextLMDataBunch.load(path='.', bs=32)\ndata.show_batch()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"759c108a6a021b85b2cbc2af9fd58a6a49bca02e"},"cell_type":"code","source":"gc.collect()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"fbcc2b77c98a430d9df8ad92d172c5e21fa352a1"},"cell_type":"code","source":"learner = language_model_learner(data, drop_mult=0.7, pretrained_model=URLs.WT103) #emb_sz=EMBED_SIZE","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"cd6edb689cbc14f4064e73becaf9e509b359821a"},"cell_type":"code","source":"#learner.lr_find()\n#learner.recorder.plot(skip_start=25)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"cc54faafe5c37dc8cace486eee680f48a1c5755b"},"cell_type":"code","source":"learner.fit_one_cycle(1, 5e-2, moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"0502ea245a311fec1e34ca4ffd5d885cd53a24db"},"cell_type":"code","source":"learner.unfreeze()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"e7b7c2b3a9d2ab429de5daa8023b4532d1a9cf3b"},"cell_type":"code","source":"learner.fit_one_cycle(1, 1e-2, moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"386e0397a47bd043d0be0d01ddf1782e4309cde1"},"cell_type":"code","source":"learner.save_encoder('fine_tuned_enc')","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"03e97e47f6ccfcde3c92a7a619fba0fdd6c5d53e"},"cell_type":"markdown","source":"### Classifier"},{"metadata":{"trusted":true,"_uuid":"3f86fd068919a9e1c2f56720a0f64009b230b232"},"cell_type":"code","source":"data = TextClasDataBunch.load(path='.', bs=32)\ndata.show_batch()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"f00067076e0a544d9c6f9789253f4c478d2def18"},"cell_type":"code","source":"learner = text_classifier_learner(data, drop_mult=0.3) #emb_sz=EMBED_SIZE\nlearner.load_encoder('fine_tuned_enc')\nlearner.freeze()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"df101a4b5402deb3f77e1985ab9c2ca09ebfe363"},"cell_type":"code","source":"#learner.lr_find()\n#learner.recorder.plot()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"87b9d18abd2f80b43652003e4647eede229da9db"},"cell_type":"code","source":"learner.fit_one_cycle(1, 5e-2, moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"401c2b10611310e1773e4f6d2c702f99877cdc5e"},"cell_type":"code","source":"#learner.freeze_to(-2)\n#learner.fit_one_cycle(1, slice(1e-3,1e-1), moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"aca8e0696051c3dbf93fb4a4996ea041e88247bd"},"cell_type":"code","source":"learner.unfreeze()\nlearner.fit_one_cycle(1, slice(1e-3/(2.6**4),1e-3), moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"ff27707f84a911c254c878de0fb307707d69954c"},"cell_type":"code","source":"#learner.fit_one_cycle(1, slice(1e-3/(2.6**4),1e-3), moms=(0.8,0.7))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"96c78ded8d5d9a6fc5c7361e813de9253ed0f094"},"cell_type":"code","source":"preds, targets = learner.get_preds()\n\npredictions = np.argmax(preds, axis = 1)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"4717e0b611c30890b1d7d29d610372acd1976093"},"cell_type":"code","source":"%matplotlib inline\nfrom sklearn import metrics\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nsns.set(font_scale=2)\n#predictions = model.predict(X_test, batch_size=1000)\n\nLABELS = ['Normal','Insincere'] \n\nconfusion_matrix = metrics.confusion_matrix(targets, predictions)\n\nplt.figure(figsize=(5, 5))\nsns.heatmap(confusion_matrix, xticklabels=LABELS, yticklabels=LABELS, annot=True, fmt=\"d\", annot_kws={\"size\": 20});\nplt.title(\"Confusion matrix\", fontsize=20)\nplt.ylabel('True label', fontsize=20)\nplt.xlabel('Predicted label', fontsize=20)\nplt.show()","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"e98558c4401ce04b1169f678fd8a947ab902f7b9"},"cell_type":"markdown","source":"### Test set"},{"metadata":{"trusted":true,"_uuid":"b07b812012da492c5aacac22cdd9ac6cafdb8c7b"},"cell_type":"code","source":"%time learner.predict(\"How much does a tutor earn in Bangalore?\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"0a7274f8dadb0d6e6e4ce57bcf8342189d6287d5"},"cell_type":"code","source":"#preds = learner.get_preds(ds_type=DatasetType.Test)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"b77950f8046e7f312fb9add079c9757526ad88d6"},"cell_type":"code","source":"#preds = preds[0].argmax(dim=1)\n#preds.sum()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"b9af1c632c62469e3927364cccec50d7e2a26150"},"cell_type":"code","source":"test_df = pd.read_csv('../input/test.csv')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"e57f3df7d3391cfe43b8058670241dcd23bb9ee3"},"cell_type":"code","source":"#test_df.drop(['question_text'], axis=1, inplace=True)\n#test_df['prediction'] = preds.numpy()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"e800e07519b839cd148e1240b6f2ca90dc7f94a4"},"cell_type":"code","source":"#test_df.to_csv(\"submission.csv\", index=False)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"d9a0ecf83cdbb0e180b216c5c285ad33867103e6"},"cell_type":"code","source":"probs, _ = learner.get_preds(DatasetType.Test)\npreds = np.argmax(probs, axis=1)\n\nsubmission = pd.DataFrame(test_df['qid'])\nsubmission['prediction'] = preds \nsubmission.to_csv('submission.csv',index=False)\nsubmission.head()","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"name":"python","version":"3.6.6","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"}},"nbformat":4,"nbformat_minor":1}