In [3]:

    

from cntk.layers import Dense
from cntk.models import Sequential
import cntk.ops as C
from cntk.ops import element_times, constant
from cntk.io import MinibatchSource, CTFDeserializer, StreamDef, StreamDefs, INFINITELY_REPEAT
from cntk.learner import sgd
from cntk import Trainer
import numpy as np


    

In [4]:

    

input_dim = 10
input = C.input_variable(input_dim, np.float32)


    

In [5]:

    

output = element_times(constant(2.), input)
x = np.asarray(np.random.uniform(size=(10, )), dtype=np.float32)
y = output.eval({input:x})
print (x)
print (y)


    

    




[ 0.7893725   0.7088815   0.11684945  0.68435621  0.51292056  0.97292995
  0.91700727  0.38358605  0.00213148  0.32982144]
[[[ 1.57874501  1.41776299  0.2336989   1.36871243  1.02584112  1.94585991
    1.83401453  0.7671721   0.00426297  0.65964288]]]

In [6]:

    

def create_reader(path, is_training, input_dim, label_dim):
    return MinibatchSource(CTFDeserializer(path, StreamDefs(
        features  = StreamDef(field='features', shape=input_dim, is_sparse=False),
        labels    = StreamDef(field='labels',   shape=label_dim, is_sparse=False)
    )), randomize=is_training, epoch_size = INFINITELY_REPEAT if is_training else FULL_DATA_SWEEP)


    

In [7]:

    

from cntk.ops import cross_entropy_with_softmax, classification_error
def simple_mnist(debug_output=False):
    input_dim = 784
    num_output_classes = 10
    num_hidden_layers = 1
    hidden_layer_dim = 200
    
    input = C.input_variable(input_dim, np.float32)
    label = C.input_variable(num_output_classes, np.float32)
    scaled_input = element_times(constant(0.00390625), input)
    model = Sequential([Dense(hidden_layer_dim, activation=C.relu),
                        Dense(num_output_classes)])
    netout = model(scaled_input)
    cross_entropy = cross_entropy_with_softmax(netout, label)
    error = classification_error(netout, label)
    trainer = Trainer(model=netout, loss_function=cross_entropy, eval_function=error,
                            parameter_learners=sgd(netout.parameters, lr=0.0001))
    #trainer = Trainer(netout, cross_entropy, error, sgd(netout.parameters, lr=0.003125))
    
    data_file = '/home/jian/repos/cntk/Examples/Image/Datasets/MNIST/Train-28x28_cntk_text.txt'
    reader_train = create_reader(data_file, True, input_dim, num_output_classes)
    input_map = {
        input: reader_train.streams.features,
        label: reader_train.streams.labels
    }


    

In [8]:

    

simple_mnist()


    

In [ ]: