tensorflow_code

images_placeholder = tf.placeholder(tf.float32, [None, training_images.shape[1], training_images.shape[2], 1], name = "input_node")
one_hot_placeholder = tf.placeholder(tf.float32, [None, 3], name = "one_hot_node")
features_placeholder = tf.placeholder(tf.float32,[None, 4], name = "features_node")
total_placeholder = tf.placeholder(tf.float32,[None,7], name = "total_node")

def convolution_layer(input, filter_size, stride, filters_number, reg_value, layer_number):

    layer_name = "convolution" + str(layer_number)

    conv_output = tf.layers.conv2d(
    input,
    filters_number,
    filter_size,
    strides=stride,
    padding='valid',
    data_format='channels_last',
    dilation_rate=(1, 1),
    activation=tf.nn.relu,
    use_bias=True,
    kernel_initializer=tf.contrib.layers.xavier_initializer(),
    bias_initializer=tf.zeros_initializer(),
    kernel_regularizer=tf.contrib.layers.l2_regularizer(reg_value),
    bias_regularizer=None,
    activity_regularizer=None,
    kernel_constraint=None,
    bias_constraint=None,
    trainable=True,
    name=layer_name,
    reuse=None)

    return conv_output

def fully_connected_layers(input, neurons_number, layer_number):
    layer_name = "fully_connected" + str(layer_number)
    full_con_output = tf.layers.dense(
        inputs=input,
        units = neurons_number,
        activation = tf.nn.relu,
        kernel_initializer = tf.contrib.layers.xavier_initializer(),
        kernel_regularizer = tf.contrib.layers.l2_regularizer(0.001),
        name=layer_name)

    return full_con_output

#architecture here
with tf.name_scope('Conv1'):
    conv1 = convolution_layer( images_placeholder, 5, 2, 6, 0.001, 1 )
    pool1 = tf.nn.max_pool(conv1, [1,2,2,1], [1,2,2,1], 'SAME', data_format='NHWC', name = "pooling_1")
with tf.name_scope('Conv2'):
    conv2 = convolution_layer( pool1, 11, 2, 6, 0.001, 2 )
    pool2 = tf.nn.max_pool(conv2, [1,2,2,1], [1,2,2,1], 'SAME', data_format='NHWC', name = "pooling_2")
    shape = pool2.get_shape()
    flatten = tf.reshape( pool2, [ -1, 216 ] )
with tf.name_scope('Dense2'):
    dense2 = fully_connected_layers( flatten, 40, 4 )
with tf.name_scope('Dense4'):
    dense4 = fully_connected_layers( dense2, 4, 6 )
    features_output = tf.nn.relu( dense4, "output_node" )