model = regression_model(
features.shape[1],
emb_mat,
seqs.shape[1],
conv_layers=config['conv_layers'],
filters=config['conv_filters'],
dropout=config['dropout'],
fc_layers=config['fc_layers'],
fc_units=config['fc_units'],
metrics=[r2],
)
save_architecture(model, model_path + '.json')
# load model callbacks
cbs = get_callbacks(model_name=model_name, log_dir=logging_path, stop_patience=10, lr_patience=4, verbose=1, emb_freq=5, emb_layers=['word_embedding'], emb_meta={'word_embedding': 'word_labels.tsv'})
# train model
model.fit(
{'text_input': seqs[:train_size], 'aux_input': features[:train_size]},
{'output': labels[:train_size]},
validation_data=({'text_input': seqs[-val_size:], 'aux_input': features[-val_size:]}, {'output': labels[-val_size:]}),
batch_size=batch_size,
epochs=200,
verbose=0,
shuffle=True,
callbacks=cbs,
)
history = cbs[2]
print_regression_metrics(history)
print("Validation set: {} examples".format(val_size))
# Create logging directory
get_ipython().system('mkdir -p $logging_path')
# Remove prior logs
get_ipython().system('rm $logging_path/*')
# load and save model
model = regression_model(feats.shape[1],
config['num_layers'],
config['num_units'],
metrics=[r2])
save_architecture(model, model_path + '.json')
# load model callbacks
cbs = get_callbacks(model_name=model_name, log_dir=logging_path, verbose=1)
# train model
model.fit(feats[:train_size],
labels[:train_size],
validation_data=(feats[-val_size:], labels[-val_size:]),
batch_size=batch_size,
epochs=100,
verbose=0,
shuffle=True,
callbacks=cbs)
# print best result
history = cbs[2]
print_regression_metrics(history)