Simple 1D (and shallow 2D) convnet and model inspection example

In [1]:

    

# Import libraries


    

In [2]:

    

# Load data


    

In [3]:

    

# Format data into 1D (single variable) long 2D arrays


    

In [4]:

    

# Conv net Model spec

def max_pool_2x1(tensor_in):
    return tf.nn.max_pool(tensor_in, ksize=[1, 2, 1, 1], strides=[1, 2, 1, 1],
        padding='SAME')

def conv_model(X, y):
    # reshape X to 4d tensor with 2nd and 3rd dimensions being image width and height
    # final dimension being the number of color channels
    X = tf.reshape(X, [-1, 100, 1, 1])
    # first conv layer will compute 32 features for each 5x1 strip
    with tf.variable_scope('conv_layer1'):
        h_conv1 = skflow.ops.conv2d(X, n_filters=32, filter_shape=[5, 1], 
                                    bias=True, activation=tf.nn.relu)
        h_pool1 = max_pool_2x1(h_conv1)
    # second conv layer will compute 64 features for each 5x1 strip
    with tf.variable_scope('conv_layer2'):
        h_conv2 = skflow.ops.conv2d(h_pool1, n_filters=64, filter_shape=[5, 1], 
                                    bias=True, activation=tf.nn.relu)
        h_pool2 = max_pool_2x1(h_conv2)
        # reshape tensor into a batch of vectors
        h_pool2_flat = tf.reshape(h_pool2, [-1, 5 * 5 * 64])
    # densely connected layer with 1024 neurons
    h_fc1 = skflow.ops.dnn(h_pool2_flat, [1024], activation=tf.nn.relu, dropout=0.5)
    return skflow.models.logistic_regression(h_fc1, y)

# # Training and predicting
# classifier3 = skflow.TensorFlowEstimator(
#     model_fn=conv_model, n_classes=10, batch_size=100, steps=20000,
#     learning_rate=0.001)


    

In [ ]:

    

# Model fit - once and compare to linear regression
# Convnet
convnet = skflow.TensorFlowEstimator(
    model_fn=conv_model, n_classes=10, batch_size=100, steps=20000,
    learning_rate=0.001,verbose=0)

convnet.fit(traindata,trainlabs1D-1)
# lr.fit(both_c,tt_c.squeeze())
# print(accuracy_score(testlabs1D-1,convnet.predict(testdata)))


    

In [5]:

    

scores_convnet = cross_validation.cross_val_score(convnet, both_c, tt_c.squeeze()-1, cv=5,scoring='accuracy') #'f1_weighted')

print(scores_convnet)
print(np.mean(scores_convnet))


    

In [6]:

    

# Weight inspection


    

In [ ]: