def adverse_validation(dev, batchez, glove):
samples = len(batchez[1])
discriminator = am.discriminator(glove, 50)
ad_model = am.adverse_model(discriminator)
res = ad_model.fit([dev[1][:samples], batchez[1]], np.zeros(samples), validation_split=0.1,
verbose = 0, nb_epoch = 20, callbacks = [EarlyStopping(patience=2)])
return np.min(res.history['val_loss'])
def adverse_model_train(model_dir, train, aug_train, dev, aug_dev, dim, glove):
discriminator = am.discriminator(glove, dim)
ad_model = am.adverse_model(discriminator)
dev_len = len(aug_dev[1])
res = ad_model.fit([train[1], aug_train[1]], np.zeros(len(train[1])),
validation_data=([dev[1][:dev_len], aug_dev[1]], np.zeros(dev_len)),
verbose = 1, nb_epoch = 5)
discriminator.save_weights(model_dir + '/adverse.weights')
def adverse_validation(dev, batchez, glove):
samples = len(batchez[1])
discriminator = am.discriminator(glove, 50)
ad_model = am.adverse_model(discriminator)
res = ad_model.fit([dev[1][:samples], batchez[1]],
np.zeros(samples),
validation_split=0.1,
verbose=0,
nb_epoch=20,
callbacks=[EarlyStopping(patience=2)])
return np.min(res.history['val_loss'])