str(Data.train[1].shape))
log.info("Validation shape: X:%s Y:%s", str(Data.valid[0].shape),
str(Data.valid[1].shape))
log.info("Testing shape: X:%s Y:%s", str(Data.test[0].shape),
str(Data.test[1].shape))
if lstnet_init.plot == True and lstnet_init.autocorrelation is not None:
AutoCorrelationPlot(Data, lstnet_init)
# If --load is set, load model from file, otherwise create model
if lstnet_init.load is not None:
log.info("Load model from %s", lstnet_init.load)
lstnet = LoadModel(lstnet_init.load, custom_objects)
else:
log.info("Creating model")
lstnet = LSTNetModel(lstnet_init, Data.train[0].shape)
if lstnet is None:
log.critical("Model could not be loaded or created ... exiting!!")
exit(1)
# Compile model
lstnet_tensorboard = ModelCompile(lstnet, lstnet_init)
if lstnet_tensorboard is not None:
log.info(
"Model compiled ... Open tensorboard in order to visualise it!")
else:
log.info(
"Model compiled ... No tensorboard visualisation is available")
# Model Training
def optimize_model(hyperparameters):
global ITERATION, lstnet_init, Data
ITERATION += 1
lstnet_init.lr = hyperparameters['lr']
lstnet_init.dropout = round(hyperparameters['dropout'], 1)
lstnet_init.GRUUnits = int(hyperparameters['GRUUnits'])
lstnet_init.batchsize = int(hyperparameters['batchsize'])
log.info("Hyper-parameters: Learning rate:%f, Batch_size:%f, Dropout:%f, GRUUnits:%f",
lstnet_init.lr,
lstnet_init.batchsize,
lstnet_init.dropout,
lstnet_init.GRUUnits)
log.info("Creating model")
lstnet = LSTNetModel(lstnet_init, Data.train[0].shape)
log.info("Compile model")
lstnet_tensorboard = ModelCompile(lstnet, lstnet_init)
if lstnet_tensorboard is not None:
log.info("Model compiled ... Open tensorboard in order to visualise it!")
else:
log.info("Model compiled ... No tensorboard visualisation is available")
log.info("Training model ... ")
start_time = datetime.now()
h = train(lstnet, Data, lstnet_init, lstnet_init.batchsize, lstnet_init.epochs, lstnet_tensorboard)
time = datetime.now() - start_time
loss, rse, corr, nrmse, nd = lstnet.evaluate(Data.valid[0], Data.valid[1]) #
log.info("Validation on the validation set returned: Loss:%f, RSE:%f, Correlation:%f, NRMSE:%f, ND:%f", loss, rse, corr, nrmse, nd)
# Write to the csv file
of_connection = open(out_file, 'a')
writer = csv.writer(of_connection)
for i in range(1, len(h.history['loss'])+1): #(1,lstnet_init.epochs+1):
train_metric = {'train_mae': h.history['loss'][i-1],
'train_rse': h.history['rse'][i-1],
'train_corr': h.history['corr'][i-1],
'train_nrmse': h.history['nrmse'][i-1],
'train_nd': h.history['nd'][i-1]}
test_metric = {'test_mae': h.history['val_loss'][i-1],
'test_rse': h.history['val_rse'][i-1],
'test_corr': h.history['val_corr'][i-1],
'test_nrmse': h.history['val_nrmse'][i-1],
'test_nd': h.history['val_nd'][i-1]}
writer.writerow([ITERATION, i, train_metric, test_metric, hyperparameters, time])
of_connection.close()
eval_time = str(datetime.now()-start_time)
if (nd == np.inf):
nd = 1.
elif (nd == np.nan):
nd = 1.
elif nd == 'nan':
nd = 1.
result = {'loss': nd,
'rse':rse,
'corr':corr,
'rse':nrmse,
'nd':nd,
'val_loss': loss,
'space': hyperparameters,
'iteration': ITERATION,
'eval_time': eval_time,
'status': STATUS_OK}
return result