best version 1.1: voting on model-5.h5

hw5/best.py

@@ -9,6 +9,9 @@
 from sklearn.multiclass import OneVsRestClassifier
 from sklearn.metrics import f1_score
 from sklearn.model_selection import cross_val_score
+from keras.models import load_model
+from keras import backend as K
+from keras.preprocessing.sequence import pad_sequences
 
 
 def read_data(file):
@@ -41,6 +44,25 @@ def read_test(file):
  return texts
 
 
+def preprocess(train, test, tokenizer, maxlen=400, split=" ", filters='!"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\'\t\n'):
+ sequences_train = tokenizer.texts_to_sequences(train)
+ sequences_test = tokenizer.texts_to_sequences(test)
+ return pad_sequences(sequences_train, maxlen=maxlen), pad_sequences(sequences_test, maxlen=maxlen)
+
+
+def fmeasure(y_true, y_pred):
+ thres = 0.4
+ y_pred = K.cast(K.greater(y_pred, thres), dtype='float32')
+ true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)), axis=-1)
+ predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)), axis=-1)
+ possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)), axis=-1)
+ precision = true_positives / (predicted_positives + K.epsilon())
+ recall = true_positives / (possible_positives + K.epsilon())
+ f = 2 * (precision * recall) / (precision + recall + K.epsilon())
+
+ return K.mean(f)
+
+
 def ensure_dir(file_path):
  directory = os.path.dirname(file_path)
  if len(directory) == 0: return
@@ -54,6 +76,7 @@ def main():
  test_texts = read_test(test_path)
  all_corpus = texts + test_texts
 
+
  ### load the vectorizer + transformer + linerSVC
  with open(vectorizer_name, 'rb') as f:
  vectorizer = pickle.load(f)
@@ -73,11 +96,48 @@ def main():
  predict = linear_svc.predict(test_data)
  # print(mlb.classes_)
 
+
+ ### load tfidf vectorizer + linerSVC
+ with open(vectorizer_name2, 'rb') as f:
+ vectorizer = pickle.load(f)
+ with open(linear_svc_name2, 'rb') as f:
+ linear_svc = pickle.load(f)
+ sequences = vectorizer.transform(texts)
+ test_data = vectorizer.transform(test_texts)
+ x_train, y_train = sequences, tags
+
+ ### cross validation
+ scores = cross_val_score(linear_svc, x_train, y_train, cv=8, scoring='f1_samples', n_jobs=-1)
+ print(scores, scores.mean(), scores.std())
+
+ ### predict
+ predict2 = linear_svc.predict(test_data)
+
+
+ ### RNN 1
+ print('load tokenizer')
+ tokenizer = pickle.load(open(tokenizer_name, 'rb'))
+ sequences, sequences_test = preprocess(texts, test_texts, tokenizer)
+ model = load_model(model_name, custom_objects={'fmeasure': fmeasure})
+ predict3 = model.predict(sequences_test)
+ predict3[predict3 < threshold] = 0
+ predict3[predict3 >= threshold] = 1
+
+
+ ### Voting
+ pred = predict + predict2 + predict3
+ pred[pred < 1.5] = 0
+ pred[pred >= 1.5] = 1
+
+
  # Test data
  ensure_dir(output_path)
  result = []
- for i, categories in enumerate(mlb.inverse_transform(predict)):
+ mlb_backup = mlb.inverse_transform(predict3)
+ for i, categories in enumerate(mlb.inverse_transform(pred)):
  ret = []
+ if len(categories) == 0:
+ categories = mlb_backup[i]
  for category in categories:
  ret.append(category)
  result.append('"{0}","{1}"'.format(i, " ".join(ret)))
@@ -102,8 +162,14 @@ def main():
 
  base_dir = './model'
  max_features_size = 40000
+
  vectorizer_name = os.path.join(base_dir, 'vec')
  transformer_name = os.path.join(base_dir, 'trans')
  linear_svc_name = os.path.join(base_dir, 'linSVC')
+ vectorizer_name2 = os.path.join(base_dir, 'vec2')
+ linear_svc_name2 = os.path.join(base_dir, 'linSVC2')
+ tokenizer_name = os.path.join(base_dir, 'word_index')
+ model_name = os.path.join(base_dir, 'model-5.h5')
+ threshold = 0.4
 
  main()