logging.debug('=' * 20)
logging.debug('生成CNN深度特征器')
from deep_learning.cnn.bow_cnn.single_channel_bow_cnn_model import SingleChannelBowCNN
bow_cnn = SingleChannelBowCNN(
rand_seed=rand_seed,
verbose=1,
feature_encoder=feature_encoder,
num_labels=len(index_to_label),
input_length=feature_encoder.vocabulary_size,
l1_conv_filter_type=[
[layer1, l1_conv_filter_type[0], -1, 'valid', (k[0], 1), 0.5],
[layer1, l1_conv_filter_type[1], -1, 'valid', (k[0], 1), 0.],
[layer1, l1_conv_filter_type[2], -1, 'valid', (k[0], 1), 0.],
],
l2_conv_filter_type=[
[layer2, l2_conv_filter_type[0], -1, 'valid', (k[1], 1), 0.5]
],
full_connected_layer_units=[hidden1, hidden2],
output_dropout_rate=0.2,
nb_epoch=nb_epoch,
earlyStoping_patience=50,
optimizers='sgd',
batch_size=32,
lr=1e-2,
)
bow_cnn.print_model_descibe()
print(bow_cnn.fit(
(train_X_feature, train_y),
(test_X_feature, test_y)))
print('-' * 80)
print('第%d个验证' % counter)
fout.write('-' * 80+'\n')
fout.write('第%d个验证\n' % counter)
bow_cnn = SingleChannelBowCNN(
rand_seed=rand_seed,
verbose=verbose,
feature_encoder=feature_encoder,
num_labels=len(index_to_label),
input_length=feature_encoder.vocabulary_size,
l1_conv_filter_type=[
[layer1, l1_conv_filter_type[0], -1, 'valid', (k[0], 1), 0.5],
[layer1, l1_conv_filter_type[1], -1, 'valid', (k[0], 1), 0.],
[layer1, l1_conv_filter_type[2], -1, 'valid', (k[0], 1), 0.],
],
l2_conv_filter_type=[
[layer2, l2_conv_filter_type[0], -1, 'valid', (k[1], 1), 0.5]
],
full_connected_layer_units=[hidden1, hidden2],
output_dropout_rate=0.2,
nb_epoch=nb_epoch,
earlyStoping_patience=50,
optimizers='sgd',
batch_size=32,
lr=1e-2,
)
dev_loss, dev_accuracy, \
val_loss, val_accuracy = bow_cnn.fit((dev_X, dev_y),(val_X, val_y))
print('dev:%f,%f'%(dev_loss, dev_accuracy))
print('val:%f,%f'%(val_loss, val_accuracy))
cv_data = data_util.get_k_fold_data(k=3,
data=train_data,
rand_seed=0,
)
SingleChannelBowCNN.cross_validation(
cv_data,
(test_data[u'SENTENCE'].as_matrix(), test_y),
'single_%s_bow_cv_detail.txt',
rand_seed=1337,
nb_epoch=30,
verbose=0,
feature_type='word_seg',
layer1=[3,5,8,18],
l1_conv_filter_type=[2, 3, 4],
layer2=[3, 7],
l2_conv_filter_type=[5],
k=[2, 2],
hidden1=[50, 100],
hidden2=[50, 100],
)
quit()
for layer1 in [3,5,8,18]:
for layer2 in [3,7,10,20,50]:
for hidden1 in [50,100,500,1000]:
for hidden2 in [50,100,300,450]:
print('=' * 150)
cv_data = data_util.get_k_fold_data(k=3,
data=train_data,
rand_seed=3,
)
SingleChannelBowCNN.cross_validation(
cv_data,
(test_data[u'SENTENCE'].as_matrix(), test_y),
'result/cnn_bow_%s_v2.3S_cv_detail.txt'% feature_type,
rand_seed=rand_seed,
nb_epoch=nb_epoch,
verbose=verbose,
remove_stopword = remove_stopword,
feature_type=feature_type,
layer1=layer1,
l1_conv_filter_type=l1_conv_filter_type,
layer2=layer2,
l2_conv_filter_type=l2_conv_filter_type,
k=k,
hidden1=hidden1,
hidden2=hidden2,
word2vec_to_solve_oov = word2vec_to_solve_oov,
word2vec_model_file_path = config['word2vec_model_file_path']
)
end_time = timeit.default_timer()
print 'end! Running time:%ds!' % (end_time - start_time)
logging.debug('=' * 20)
**config)
train_X_feature, train_y, test_X_feature, test_y,_,_ = all_cv_data[0]
bow_cnn = SingleChannelBowCNN(
rand_seed=rand_seed,
verbose=1,
feature_encoder=feature_encoder,
num_labels=len(index_to_label),
input_length=train_X_feature.shape[-1],
l1_conv_filter_type=[
# [layer1, l1_conv_filter_type[0], -1, 'valid', (k[0], 1), 0., 'relu', 'none'],
# [layer1, l1_conv_filter_type[1], -1, 'valid', (k[0], 1), 0., 'relu', 'batch_normalization'],
# [layer1, l1_conv_filter_type[2], -1, 'valid', (k[0], 1), 0., 'relu', 'batch_normalization'],
],
l2_conv_filter_type=[
[layer2, -1, -1, 'valid', (0, 1), 0., 'relu', 'none'],
# [layer2, l2_conv_filter_type[1], -1, 'valid', (k[1], 1), 0.5]
],
full_connected_layer_units=[
[hidden1, 0., 'relu', 'none'],
# [hidden2, 0.5, 'relu', 'none']
],
nb_epoch=nb_epoch,
earlyStoping_patience=50,
optimizers='sgd',
batch_size=32,
lr=2e-2,
)
bow_cnn.print_model_descibe()
print(bow_cnn.fit(