else:
save_path = args.save_dir + '/cnn_s2p_' + appliance_name + '_pointnet_model'
if not os.path.exists(save_path):
os.makedirs(save_path)
# Calling custom training function
train_loss, val_loss, step_train_loss, step_val_loss = nf.customfit(
sess=sess,
network=model,
cost=cost,
train_op=train_op,
tra_provider=tra_provider,
x=x,
y_=y_,
acc=None,
n_epoch=args.n_epoch,
print_freq=1,
val_provider=val_provider,
save_model=args.save_model,
save_path=save_path,
epoch_identifier=None,
earlystopping=True,
min_epoch=1,
patience=10)
# Following are training info
log('train loss: ' + str(train_loss))
log('val loss: ' + str(val_loss))
# infos = pd.DataFrame(data={'train_loss': step_train_loss,
# #'val_loss': step_val_loss
weights_loader(model, param_file)
# Calling custom test function
# test_prediction = nf.custompredict_add(sess=sess,
# network=model,
# output_provider = test_provider,
# x = x,
# fragment_size=args.nosOfWindows,
# output_length=windowlength,
# y_op= None,
# out_kwag=test_kwag,seqlength = test_set_x.size)
test_prediction = nf.custompredictS2SX(sess=sess,
network=model,
output_provider=test_provider,
x=x,
fragment_size=args.nosOfWindows,
output_length=windowlength,
y_op=None,
out_kwag=test_kwag)
# ------------------------------------- Performance evaluation----------------------------------------------------------
# Parameters
max_power = params_appliance[args.appliance_name]['max_on_power']
threshold = params_appliance[args.appliance_name]['on_power_threshold']
aggregate_mean = 522
aggregate_std = 814
appliance_mean = params_appliance[args.appliance_name]['mean']
appliance_std = params_appliance[args.appliance_name]['std']
name='output_layer')
y = network.outputs
param_file = 'cnn' + app2 + '_s2s_model_check.npz'
print(param_file)
#params = tl.files.load_npz(path='./', name=param_file)
params = tl.files.load_npz(path='', name=param_file)
tl.files.assign_params(sess, params, network)
print('params done')
test_prediction = nf.custompredict_add(sess=sess,
network=network,
output_provider=test_provider,
x=x,
fragment_size=window_size,
output_length=windowlength,
y_op=None,
out_kwag=test_kwag,
seqlength=test_set_x.size,
std=std,
mean=mean)
max_power = params_appliance[application]['max_on_power']
threshold = params_appliance[application]['on_power_threshold']
ground_truth = ground_truth[offset:-offset] * std + mean
mains = (test_set_x[offset:-offset]) * std + mean
prediction = test_prediction[offset:-offset]
prediction[prediction <= 0.0] = 0.0
print(prediction.shape)
use_locking=False).minimize(
cost, var_list=train_params)
# initialize all variables
sess.run(tf.global_variables_initializer())
# params = tl.files.load_npz(path='', name='cnn_lstm_model.npz')
# tl.files.assign_params(sess, params, network)
print 'set sucessful'
save_path = './cnn_' + appliance_name + '_s2p'
nf.customfit(sess=sess,
network=network,
cost=cost,
train_op=train_op,
tra_provider=tra_provider,
x=x,
y_=y_,
acc=None,
n_epoch=epoch,
print_freq=1,
val_provider=val_provider,
save_model=saver,
tra_kwag=tra_kwag,
val_kwag=val_kwag,
save_path=save_path,
epoch_identifier=None,
earlystopping=True,
min_epoch=1,
patience=10)
sess.close()
