Nolearn net to do supervised learning

gistfile1.py

This is an example nolearn NeuralNet to do supervised learning. This network is being used to do supervised learning on black and white images, which are 135 * 240 in size. The same network can be modified to bgr images just by modifying the input_shape to None,3,..,.. and output_num_unit to 3*135 * 240

net_unsupervised_bw = NeuralNet(
    layers=[
        ('input', layers.InputLayer),
        ('noise1', GaussianNoiseLayer),
        ('conv1', Conv2DLayer),
        ('pool1', MaxPool2DLayer),
        ('dropout1', layers.DropoutLayer),  # !
        ('conv2', Conv2DLayer),
        ('pool2', MaxPool2DLayer),
        ('dropout2', layers.DropoutLayer),  # !
        ('conv3', Conv2DLayer),
        ('pool3', MaxPool2DLayer),
        ('dropout3', layers.DropoutLayer),  # !
        ('hidden1', layers.DenseLayer),
        ('hidden2', layers.DenseLayer),
        ('hidden3', layers.DenseLayer),
        ('output', layers.DenseLayer),
        ],
    batch_iterator=BatchIterator(20,True),
    input_shape=(None, 1, 135, 240),
    conv1_num_filters=32, conv1_filter_size=(12, 21), pool1_ds=(2, 2),
    dropout1_p=0.1,  # !
    conv2_num_filters=64, conv2_filter_size=(7,11), pool2_ds=(2, 2),
    dropout2_p=0.2,  # !
    conv3_num_filters=128, conv3_filter_size=(5,9), pool3_ds=(2, 2),
    dropout3_p=0.3,  # !
    hidden3_num_units=100, hidden3_nonlinearity=sigmoid,
    hidden1_num_units=100, hidden1_nonlinearity=sigmoid,
    hidden2_num_units=100, hidden2_nonlinearity=sigmoid, 
    output_num_units=135*240, output_nonlinearity=None,
    on_epoch_finished=[
        #AdjustVariable('update_learning_rate', start=0.03, stop=0.0001),
        #AdjustVariable('update_momentum', start=0.9, stop=0.999),
        EarlyStopping(patience=20),
        ],    
    #updated=adagrad,
    update_learning_rate=0.03,
    update_momentum=0.95,
    regression=True,
    max_epochs=60,
    verbose=1,
    )

def load_data(dataset):
    # Load the dataset
    f = open(dataset, 'rb')
    dump = cPickle.load(f)
    train_set = dump['train']
    #valid_set = dump['validate']
    #test_set = dump['test']
    f.close()
    
    def get_dataset(data_xy, borrow=True):
        data_x, data_y = data_xy
        
        x = np.asarray(data_x,dtype=np.float32)/255.
        y = np.asarray(data_x,dtype=np.float32)/255.
        return x , y

    #test_set_x, test_set_y = get_dataset(test_set)
    #valid_set_x, valid_set_y = get_dataset(valid_set)
    train_set_x, train_set_y = get_dataset(train_set)

    rval = [(train_set_x, train_set_y)]#,(test_set_x, test_set_y)] #, (valid_set_x, valid_set_y),
    return rval

datasets = load_data(TRAINFILE)
X, y = datasets[0]

print("X.shape == {}; X.min == {:.3f}; X.max == {:.3f}".format(
    X.shape, X.min(), X.max()))
print("y.shape == {}; y.min == {:.3f}; y.max == {:.3f}".format(
    y.shape, y.min(), y.max()))

if(bw=='color'):
    X = X.reshape(-1, 3, 135, 240)
    y = y.reshape(-1, 3 * 135 * 240)
else:
    X = X.reshape(-1, 1, 135, 240)
    y = y.reshape(-1, 135 * 240)

print("X.shape == {}; X.min == {:.3f}; X.max == {:.3f}".format(
    X.shape, X.min(), X.max()))
print("y.shape == {}; y.min == {:.3f}; y.max == {:.3f}".format(
    y.shape, y.min(), y.max()))


net = net_unsupervised_bw

if(bw=='color'):
    net.input_shape=(None, 3, 135, 240)
    net.output_num_units=3* 135 * 240

net.max_epochs = int(epochs)
net.loss = mse
net.fit(X, y)