{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        print(os.path.join(dirname, filename))\n\n# You can write up to 5GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"# Load packages\n# Load packages \nimport math\nimport re\nimport os\nimport timeit\nimport tensorflow as tf\nimport numpy as np\nimport pandas as pd\nimport seaborn as sns\nimport matplotlib.pyplot as plt\nimport nltk\nimport logging\nimport time\nimport smart_open\n\nfrom sklearn import svm\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import accuracy_score, confusion_matrix, classification_report, f1_score\nfrom sklearn.naive_bayes import MultinomialNB\nfrom sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\nlogging.basicConfig(format='[%(asctime)s %(levelname)8s] %(message)s', level=logging.INFO, datefmt='%m-%d %H:%M:%S')\n\nimport keras\nfrom keras.models import Model\nfrom keras import Input, layers\n\nfrom keras.callbacks import EarlyStopping, ModelCheckpoint\nfrom keras.models import Sequential\nfrom keras.layers import Flatten, Dense, Embedding, Dropout, LSTM, GRU, Bidirectional\nfrom keras.utils import to_categorical\nfrom keras.preprocessing.text import Tokenizer\nfrom keras.preprocessing.sequence import pad_sequences\nimport gensim.downloader as api\n\n# Get data\n! cp ../input/quora-insincere-questions-classification/*.csv .","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# Base class for classifier\nclass Classifier():\n  def __init__(self):\n    self.train = None\n    self.test = None \n    self.model = None\n\n  def load_data(self, train_file='train.csv', test_file='test.csv'):\n      \"\"\" Load train, test csv files and return pandas.DataFrame\n      \"\"\"\n      train = pd.read_csv(train_file, engine='python', encoding='utf-8', error_bad_lines=False)\n      self.test = pd.read_csv(test_file, engine='python', encoding='utf-8', error_bad_lines=False)\n      self.train = pd.concat([train[train.target == 1].sample(80_000), train[train.target==0].sample(200_000)])\n      logging.info('CSV data loaded')\n  \n  def countvectorize(self):\n      tv = TfidfVectorizer(ngram_range=(1,3), token_pattern=r'\\w{1,}',\n               min_df=3, max_df=0.9, strip_accents='unicode', use_idf=1,\n               smooth_idf=1, sublinear_tf=1, max_features=5000)\n#       tv = CountVectorizer()\n      tv.fit(self.train.question_text)\n      self.vector_train = tv.transform(self.train.question_text)\n      self.vector_test  = tv.transform(self.test.question_text)\n      logging.info(\"Train & test text tokenized\")\n\n  def build_model(self):\n      pass\n\n  def run_model(self):\n      # Choose your own classifier: self.model and run it\n      logging.info(f\"{self.__class__.__name__} starts running.\")\n      labels = self.train.target\n      x_train, x_val, y_train, y_val = train_test_split(self.vector_train, labels, test_size=0.2, random_state=2090)\n      self.model.fit(x_train, y_train)\n      y_preds = self.model.predict(x_val)\n\n      logging.info(f\"Accuracy score: {accuracy_score(y_val, y_preds)}\")\n      logging.info(f\"Confusion matrix: \") \n      print(confusion_matrix(y_val, y_preds))\n      print(\"Classificaiton report:\\n\", classification_report(y_val, y_preds, target_names=[\"Sincere\", \"Insincere\"]))\n      # y_preds = self.model.predict(self.vector_test)\n      return y_preds\n\n  def save_predictions(self, y_preds):\n      sub = pd.read_csv(f\"sample_submission.csv\")\n      sub['prediction'] = y_preds \n      sub.to_csv(f\"submission_{self.__class__.__name__}.csv\", index=False)\n      logging.info('Prediction exported to submisison.csv')\n  \n  def pipeline(self):\n      s_time = time.clock()\n      self.load_data()\n      self.countvectorize()\n      self.build_model()\n      self.save_predictions(self.run_model())\n      logging.info(f\"Program running for {time.clock() - s_time} seconds\")\n\nclass C_Bayes(Classifier):\n  def build_model(self):\n      self.model = MultinomialNB()\n      return self.model\n\n# Logistic Regression \nclass C_LR(Classifier):\n  def build_model(self):\n      self.model = LogisticRegression(n_jobs=10, solver='lbfgs', C=0.1, verbose=1)\n      return self.model\n\nclass C_SVM(Classifier):\n  def load_data(self, train_file='train.csv', test_file='test.csv'):\n      \"\"\" Load train, test csv files and return pandas.DataFrame\n      \"\"\"\n      self.train = pd.read_csv(train_file, engine='python', encoding='utf-8', error_bad_lines=False)\n      self.train = self.train.sample(100000)\n      self.test = pd.read_csv(test_file, engine='python', encoding='utf-8', error_bad_lines=False)\n      logging.info('CSV data loaded')\n\n  def build_model(self):\n      self.model = svm.SVC()\n      return self.model\n\nclass C_Ensemble(Classifier):\n  def ensemble(self):\n      s_time = time.perf_counter()\n      self.load_data()\n      self.countvectorize()\n\n      nb = MultinomialNB()\n      lr = LogisticRegression(n_jobs=10, solver='saga', C=0.1, verbose=1)\n      svc = svm.SVC()\n\n      all_preds = [0] * self.test.shape[0]\n      for m in (nb, lr, svc):\n          self.model = m\n          if m == svc: \n              self.load_data()\n              self.train = self.train.sample(10000)\n              self.countvectorize()\n          all_preds += self.run_model()\n\n      all_preds = [1 if p > 0 else 0 for p in all_preds]\n      self.save_predictions(all_preds)\n      logging.info(f\"Program running for {time.perf_counter() - s_time} seconds\")\n\n\nclass Helper():\n    def locate_threshold(self, model, x_val, y_val):\n        y_probs = model.predict(x_val, batch_size=1024, verbose=1)\n        best_threshold = best_f1 = pre_f1 = 0\n        history = []\n\n        for i in np.arange(0.01, 1, 0.01):\n          if len(y_probs[0]) >= 2:\n              y2_preds = [1 if e[1] >= i else 0 for e in y_probs]\n          else:\n              y2_preds = (y_probs > i).astype(int)\n\n          cur_f1 = f1_score(y_val, y2_preds)\n          history.append((i, cur_f1))\n          symbol = '+' if cur_f1 >= pre_f1 else '-'\n          print(\"Threshold {:6.4f}, f1_score: {:<0.8f}  {} {:<0.6f} \".format(i, cur_f1, symbol, abs(cur_f1 - pre_f1)))\n          pre_f1 = cur_f1\n\n          if cur_f1 >= best_f1:\n              best_f1 = cur_f1\n              best_threshold = i\n\n        print(f\"Best f1 score {best_f1}, best threshold {best_threshold}\")\n        plt.xlabel('Threshold')\n        plt.ylabel('f1_score')\n        plt.plot(*zip(*history))\n\n        return best_threshold","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class C_NN(Classifier):\n    def __init__(self, max_features=100000, embed_size=128, max_len=300):\n        self.max_features=max_features\n        self.embed_size=embed_size\n        self.max_len=max_len\n    \n    def tokenize_text(self, text_train, text_test):\n        '''@para: max_features, the most commenly used words in data set\n        @input are vector of text\n        '''\n        tokenizer = Tokenizer(num_words=self.max_features)\n        text = pd.concat([text_train, text_test])\n        tokenizer.fit_on_texts(text)\n\n        sequence_train = tokenizer.texts_to_sequences(text_train)\n        tokenized_train = pad_sequences(sequence_train, maxlen=self.max_len)\n        logging.info('Train text tokeninzed')\n\n        sequence_test = tokenizer.texts_to_sequences(text_test)\n        tokenized_test = pad_sequences(sequence_test, maxlen=self.max_len)\n        print('Test text tokeninzed')\n        return tokenized_train, tokenized_test, tokenizer\n      \n    def build_model(self):\n        dropout = 0.2\n        model = Sequential()\n        model.add(Embedding(self.max_features, self.embed_size, input_length=self.max_len))\n        model.add(Bidirectional(GRU(64, return_sequences=True)))\n        model.add(Bidirectional(GRU(64, return_sequences=True)))\n\n        model.add(Flatten())\n\n        model.add(Dense(32, activation='relu'))\n        model.add(Dropout(dropout))\n        model.add(Dense(32, activation='relu'))\n        model.add(Dropout(dropout))\n        \n        model.add(Dense(1, activation='sigmoid'))\n        self.model = model\n\n        return self.model\n\n    def run(self, x_train, y_train):\n        checkpoint = ModelCheckpoint('weights_base_best.hdf5', monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n        early = EarlyStopping(monitor=\"val_acc\", mode=\"max\", patience=5)\n\n        self.model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc'])\n        X_tra, X_val, y_tra, y_val = train_test_split(x_train, y_train, train_size=0.8, random_state=2020)\n        BATCH_SIZE = max(16, 2 ** int(math.log(len(X_tra) / 100, 2)))\n        print(f\"Batch size is set to {BATCH_SIZE}\")\n        history = self.model.fit(X_tra, y_tra, epochs=2, batch_size=BATCH_SIZE, validation_data=(X_val, y_val), \\\n                              callbacks=[checkpoint, early], verbose=1)\n\n        y_pred = self.model.predict(X_val, batch_size=64, verbose=1)\n        y_pred_bool = np.argmax(y_pred, axis=1)\n        print(classification_report(y_val, y_pred_bool))\n        return history\n\n    \n# c = C_NN(max_features=50000, embed_size=300, max_len=250)\n# c.load_data()\n# # c.train = c.train.sample(10000)\n# # c.test = c.test.sample(1000)\n# vector_train, vector_test, _ = c.tokenize_text(c.train.question_text, c.test.question_text)\n# model = c.build_model()\n# print(\"DONE\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class Kerasapi():\n    def __init__(self):\n        self.embed_size=300\n        self.max_features=50000\n        self.max_len=100\n    \n    def build_model(self):\n        text_input = Input(shape=(self.max_len, ))\n        embed_text = layers.Embedding(self.max_features, self.embed_size)(text_input)\n        \n        branch_a = layers.Bidirectional(layers.GRU(64, return_sequences=True))(embed_text)\n        branch_b = layers.GlobalMaxPool1D()(branch_a)\n        branch_c = layers.Dense(64, activation='relu')(branch_b)\n        branch_d = layers.Dropout(0.3)(branch_c)\n        branch_c = layers.Dense(64, activation='relu')(branch_b)\n        branch_d = layers.Dropout(0.3)(branch_c)\n        branch_z = layers.Dense(1, activation='sigmoid')(branch_d)\n        \n        model = Model(inputs=text_input, outputs=branch_z)\n        model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])\n        \n        return model\n\nc = C_NN(max_features=50000, embed_size=300, max_len=100)\nc.load_data()\nvector_train, vector_test, _ = c.tokenize_text(c.train.question_text, c.test.question_text)\n\nmodel = Kerasapi().build_model()\nprint(\">> model was built.\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"X_tra, X_val, y_tra, y_val = train_test_split(vector_train, c.train.target, test_size=0.1, random_state=0)\nprint(\">> train test split DONE.\")\n\nmc = keras.callbacks.ModelCheckpoint('best.h5', monitor='val_accuracy', save_best_only=True, verbose=1)\nes = keras.callbacks.EarlyStopping(monitor='val_accuracy', patience=5, verbose=1)\n\nhistory = model.fit(X_tra, y_tra, epochs=30, batch_size=1024, callbacks=[mc, es], validation_data=(X_val, y_val), verbose=1)\n\n# y_pred = c.model.predict(X_val, batch_size=1024, verbose=1)\nprint(\">> train completed.\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"model.load_weights('best.h5')\nbest_threshold = Helper().locate_threshold(model, X_val, y_val)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"sub = pd.read_csv(f\"sample_submission.csv\")\ny_preds = model.predict(vector_test, batch_size=1024, verbose=1)\ny_preds = (y_preds > best_threshold).astype(int)\nsub.prediction = y_preds\nsub.to_csv('submission.csv', index=False)","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":4}