def main():
#Get args for running
all_mesh_terms_path = ''
abstracts_path = ''
model_name = ''
n_hidden_units = 300
sample_size = 500000
try:
opts, args = getopt.getopt(sys.argv[1:], 'm:d:n:e:h:s:b:mb:p:', ['mesh_terms=', 'abstract_data','model_name',
'hidden_units', 'sample_size',])
except getopt.GetoptError:
sys.exit(2)
for opt, arg in opts:
if opt in ('-m', '--mesh_terms'):
all_mesh_terms_path = arg
elif opt in ('-d', '--abstract_data'):
abstracts_path = arg
elif opt in ('-n', '--model_name'):
model_name = arg
elif opt in ('-h', 'hidden_units'):
n_hidden_units = int(arg)
elif opt in ('-s', 'sample_size'):
sample_size = int(arg)
else:
sys.exit(2)
model = ANNMesh.create_model()
def run_train_model(all_mesh_terms_path, abstracts_path, model_name, epochs, n_hidden_units, sample_size,
batch_size, mini_batch_size, full_path):
print('..loading data')
target_dict = parse_mesh(all_mesh_terms_path, False)
X_train, Y_train = process_save_data(sample_size, target_dict, full_path, abstracts_path)
model = ANNMesh.create_model(X_train['data'].shape[1], n_hidden_units, len(target_dict))
if os.path.exists(model_name):
model.load_weights(model_name)
print('..training')
checkpointer = ModelCheckpoint(filepath=model_name, verbose=1, save_best_only=True)
loss_history = LossHistory()
model, train_loss_history, valid_loss_history = train(epochs, model, sample_size, abstracts_path, checkpointer,
loss_history, batch_size, target_dict, mini_batch_size, X_train,
Y_train)
print(len(train_loss_history), len(valid_loss_history))
print('..saving model')
model.save_weights("final_" + model_name)
#Plot the loss curves
epochs_axis = numpy.arange(1, epochs + 1)
train_loss, = plt.plot(epochs_axis, train_loss_history, label='Train')
val_loss, = plt.plot(epochs_axis, valid_loss_history, label='Validation')
plt.legend(handles=[train_loss, val_loss])
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.savefig(model_name + '_loss_plot.png')
