loglevel=args.verbosity)
loss_function = loss.SquaredErrorFunction()
gdo = optimizer.GradientDescentOptimizer(learning_rate=params.learning_rate)
nn_model = model.ModelWrapper(network=fc_net,
optimizer=gdo,
loss_function=loss_function,
x=X,
y_target=labels_onehot,
params=params)
# Run training
loss = nn_model.train()
nn_model.network.saveModel('./models', 'trained.pkl')
score_train = sup.computeScore(fc_net, X, labels_onehot)
print(
'The classification score on the training data is {}.'.format(score_train))
# Plotting training specs
if plotting_active:
pl.figure('Loss evolution')
ax = pl.gca()
ax.plot(loss[:, 0], loss[:, 1], c='b', label='training error')
ax.set_ylim([-0.2, 1.2])
ax.set_xlabel('Training step')
ax.set_ylabel('Training error')
# Visualization of classification results
mesh_step_size = 0.02
x1_min, x1_max = X[:, 0].min() - .5, X[:, 0].max() + .5
action='store_true',
default=False)
return parser.parse_args()
args = parse_args()
test_data = pkl.load(open('./data/data_test.pkl', 'rb'))
fc_net = network.FCNetwork()
fc_net.loadModel('./models/', 'trained.pkl')
X_test = test_data[0]
labels_test = test_data[1]
# Compute scores on training and testing data
score_test = sup.computeScore(fc_net, X_test, labels_test)
print('The classification score on the test data is {}'.format(score_test))
# Visualization of classification results
if args.plot:
mesh_step_size = 0.05
x1_min, x1_max = X_test[:, 0].min() - .5, X_test[:, 0].max() + .5
x2_min, x2_max = X_test[:, 1].min() - .5, X_test[:, 1].max() + .5
xx1, xx2 = np.meshgrid(np.arange(x1_min, x1_max, mesh_step_size),
np.arange(x2_min, x2_max, mesh_step_size))
predicted_label = np.zeros(xx1.shape)
for ii in range(xx1.shape[0]):
for jj in range(xx1.shape[1]):
x_query = np.array([[xx1[ii, jj], xx2[ii, jj]]])
y_net = fc_net.output(x_query)
