def fit(self, sess, h_np, b_np, h_len, b_len, y, dev_h, dev_b, dev_h_len,
dev_b_len, dev_y): #M
#losses = []
losses_epochs = [] #M
dev_performances_epochs = [] # M
dev_predictions_epochs = [] #M
dev_predicted_classes_epochs = [] #M
for epoch in range(self.config.n_epochs):
print('-------new epoch---------')
loss = self.run_epoch(sess, h_np, b_np, h_len, b_len, y)
# Computing predictions #MODIF
dev_predictions = self.predict_on_batch(sess, dev_h, dev_b,
dev_h_len, dev_b_len)
# Computing development performance #MODIF
dev_predictions = softmax(np.array(dev_predictions))
dev_predicted_classes = np.argmax(dev_predictions, axis=1)
dev_performance = get_performance(dev_predicted_classes,
dev_y,
n_classes=4)
# Adding to global outputs #MODIF
dev_predictions_epochs.append(dev_predictions)
dev_predicted_classes_epochs.append(dev_predicted_classes)
dev_performances_epochs.append(dev_performance)
losses_epochs.append(loss)
print('EPOCH: ', epoch, ', LOSS: ', np.mean(loss))
return losses_epochs, dev_performances_epochs, dev_predicted_classes_epochs, dev_predictions_epochs
def fit(self, sess, train, dev_data_np, dev_seqlen,
dev_labels): # MODIF # CAREFUL DEV/dev
'''
Returns LISTS:
- losses_epochs
- dev_performances_epochs
- dev_predictions_epochs
- dev_predicted_classes_epochs
'''
losses_epochs = [] #M
dev_performances_epochs = [] # MODIF
dev_predictions_epochs = [] #M
dev_predicted_classes_epochs = [] #M
for epoch in range(self.config.n_epochs):
logger.info("Epoch %d out of %d", epoch + 1, self.config.n_epochs)
loss = self.run_epoch(sess, train)
# Computing predictions # MODIF
dev_predictions = self.predict_on_batch(sess, dev_data_np,
dev_seqlen) #OUCH
# Computing development performance #MODIF
dev_predictions = softmax(np.array(dev_predictions))
dev_predicted_classes = np.argmax(dev_predictions, axis=1)
dev_performance = get_performance(dev_predicted_classes,
dev_labels,
n_classes=4)
# Adding to global outputs #MODIF
dev_predictions_epochs.append(dev_predictions)
dev_predicted_classes_epochs.append(dev_predicted_classes)
dev_performances_epochs.append(dev_performance)
losses_epochs.append(loss)
return losses_epochs, dev_performances_epochs, dev_predicted_classes_epochs, dev_predictions_epochs
def fit(self, sess, h_np, b_np, h_len, b_len, y, dev_h, dev_b, dev_h_len,
dev_b_len, dev_y): #M
#losses = []
losses_epochs = [] #M
dev_performances_epochs = [] # M
dev_predictions_epochs = [] #M
dev_predicted_classes_epochs = [] #M
LABELS = ['agree', 'disagree', 'discuss', 'unrelated']
for epoch in range(self.config.n_epochs):
print('-------new epoch---------')
loss = self.run_epoch(sess, h_np, b_np, h_len, b_len, y)
# Computing predictions #MODIF
dev_predictions = self.predict_on_batch(sess, dev_h, dev_b,
dev_h_len, dev_b_len)
# Computing development performance #MODIF
dev_predictions = softmax(np.array(dev_predictions))
dev_predicted_classes = np.argmax(dev_predictions, axis=1)
dev_performance = get_performance(dev_predicted_classes,
dev_y,
n_classes=4)
# Adding to global outputs #MODIF
dev_predictions_epochs.append(dev_predictions)
dev_predicted_classes_epochs.append(dev_predicted_classes)
dev_performances_epochs.append(dev_performance)
losses_epochs.append(loss)
print("=================================================")
for i, (x, y) in enumerate(zip(dev_y, dev_predicted_classes)):
print("Given class -> ", LABELS[dev_y[i]],
"Predicted class -> ", LABELS[dev_predicted_classes[i]])
print("=================================================")
print("prediction epochs: ", dev_performances_epochs)
print("predicted classes epochs: ", dev_predicted_classes_epochs)
print("performances epochs: ", dev_performances_epochs)
print('EPOCH: ', epoch, ', LOSS: ', np.mean(loss))
return losses_epochs, dev_performances_epochs, dev_predicted_classes_epochs, dev_predictions_epochs
def fit(self, sess, h_np, b_np, h_len, b_len, y, dev_h, dev_b, dev_h_len,
dev_b_len, dev_y): #M
#losses = []
data_path = '/home/neha/sem2/nlu/project/stance_detection-master/code/models/lstm_conditional/'
model_save_name = 'lstm-conditional-300blen-2l-50e'
losses_epochs = [] #M
dev_performances_epochs = [] # M
dev_predictions_epochs = [] #M
dev_predicted_classes_epochs = [] #M
saver = tf.train.Saver()
for epoch in range(self.config.n_epochs):
print('-------new epoch---------')
loss = self.run_epoch(sess, h_np, b_np, h_len, b_len, y)
# Computing predictions #MODIF
dev_predictions = self.predict_on_batch(sess, dev_h, dev_b,
dev_h_len, dev_b_len)
# Computing development performance #MODIF
dev_predictions = softmax(np.array(dev_predictions))
dev_predicted_classes = np.argmax(dev_predictions, axis=1)
dev_performance = get_performance(dev_predicted_classes,
dev_y,
n_classes=4)
# Adding to global outputs #MODIF
dev_predictions_epochs.append(dev_predictions)
dev_predicted_classes_epochs.append(dev_predicted_classes)
dev_performances_epochs.append(dev_performance)
losses_epochs.append(loss)
print('EPOCH: ', epoch, ', LOSS: ', np.mean(loss))
if epoch % 5 == 0:
saver.save(sess,
data_path + model_save_name,
global_step=epoch)
# pred_prob=pd.DataFrame(dev_predictions_epochs[1])
# pred_prob.to_csv('/home/neha/sem2/nlu/project/stance_detection-master/code/output/lstm_conditional'+'/pred_prob_' + '.csv',index = False)
saver.save(sess, data_path + model_save_name + '-final')
return losses_epochs, dev_performances_epochs, dev_predicted_classes_epochs, dev_predictions_epochs
def compute_pred(self, sess, dev_h, dev_b, dev_h_len, dev_b_len,
dev_y): #M
#losses = []
losses_epochs = [] #M
dev_performances_epochs = [] # M
dev_predictions_epochs = [] #M
dev_predicted_classes_epochs = [] #M
# for epoch in range(self.config.n_epochs):
print('ready to predict')
# loss = self.run_epoch(sess, h_np, b_np, h_len, b_len, y)
loss = [100]
# Computing predictions #MODIF
dev_predictions = self.predict_on_batch(sess, dev_h, dev_b, dev_h_len,
dev_b_len)
# Computing development performance #MODIF
dev_predictions = softmax(np.array(dev_predictions))
dev_predicted_classes = np.argmax(dev_predictions, axis=1)
dev_performance = get_performance(dev_predicted_classes,
dev_y,
n_classes=4)
# Adding to global outputs #MODIF
dev_predictions_epochs.append(dev_predictions)
dev_predicted_classes_epochs.append(dev_predicted_classes)
dev_performances_epochs.append(dev_performance)
losses_epochs.append(loss)
print('Prediction complete')
pred_prob = pd.DataFrame(dev_predictions_epochs[0])
pred_prob.to_csv(
'/home/neha/sem2/nlu/project/stance_detection-master/code/output/lstm_conditional'
+ '/pred_prob_' + '.csv',
index=False)
return losses_epochs, dev_performances_epochs, dev_predicted_classes_epochs, dev_predictions_epochs