def custom_model_layer_fully_connected(params, network_initializer):
'''
Append fc (dense) to custom network
Args:
params (dict): All layer parameters
Returns:
neural network layer: fc (dense) layer
'''
units = params["units"]
activation = None
use_bias = params["use_bias"]
kernel_initializer = get_initializer(network_initializer)
bias_initializer = 'zeros'
kernel_regularizer = None
bias_regularizer = None
activity_regularizer = None
kernel_constraint = None
bias_constraint = None
layer = keras.layers.Dense(units,
activation=activation,
use_bias=use_bias,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
kernel_regularizer=kernel_regularizer,
bias_regularizer=bias_regularizer,
activity_regularizer=activity_regularizer,
kernel_constraint=kernel_constraint,
bias_constraint=bias_constraint)
return layer
def custom_model_layer_transposed_convolution3d(params, network_initializer):
'''
Append 3d-transposed-convolution to custom network
Args:
params (dict): All layer parameters
Returns:
neural network layer: 3d-transposed-convolution layer
'''
out_channels = params["output_channels"]
kernel_size = params["kernel_size"]
strides = params["stride"]
if (params["padding"] == "in_eq_out"):
padding = "same"
elif (params["padding"] == 0):
padding = "valid"
else:
padding = "causal"
if (params["layout"][-1] == "C"):
data_format = 'channels_last'
else:
data_format = 'channels_first'
dilation_rate = params["dilation"]
activation = None
use_bias = params["use_bias"]
kernel_initializer = get_initializer(network_initializer)
bias_initializer = 'zeros'
kernel_regularizer = None
bias_regularizer = None
activity_regularizer = None
kernel_constraint = None
bias_constraint = None
layer = keras.layers.Conv3DTranspose(
out_channels,
kernel_size,
strides=strides,
padding=padding,
data_format=data_format,
dilation_rate=dilation_rate,
activation=activation,
use_bias=use_bias,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
kernel_regularizer=kernel_regularizer,
bias_regularizer=bias_regularizer,
activity_regularizer=activity_regularizer,
kernel_constraint=kernel_constraint,
bias_constraint=bias_constraint)
return layer
def custom_model_layer_convolution2d(params, network_initializer):
'''
Append 2d-convolution to custom network
Args:
params (dict): All layer parameters
Returns:
neural network layer: 2d-convolution layer
'''
out_channels = params["output_channels"];
kernel_size=params["kernel_size"];
strides=params["stride"];
if(params["padding"] == "in_eq_out"):
padding = "same";
elif(params["padding"] == 0):
padding = "valid";
else:
padding = "causal"; #causal
if(params["layout"][-1] == "C"):
data_format='channels_last';
else:
data_format='channels_first';
dilation_rate=params["dilation"];
activation=None;
use_bias = params["use_bias"];
kernel_initializer=get_initializer(network_initializer);
bias_initializer='zeros';
kernel_regularizer=None;
bias_regularizer=None;
activity_regularizer=None;
kernel_constraint=None;
bias_constraint=None;
layer = keras.layers.Conv2D(out_channels, kernel_size, strides=strides,
padding=padding, data_format=data_format,
dilation_rate=dilation_rate, activation=activation, use_bias=use_bias,
kernel_initializer=kernel_initializer, bias_initializer=bias_initializer,
kernel_regularizer=kernel_regularizer, bias_regularizer=bias_regularizer,
activity_regularizer=activity_regularizer, kernel_constraint=kernel_constraint,
bias_constraint=bias_constraint)
return layer
def custom_model_layer_convolution3d(params, network_initializer):
out_channels = params["output_channels"]
kernel_size = params["kernel_size"]
strides = params["stride"]
if (params["padding"] == "in_eq_out"):
padding = "same"
elif (params["padding"] == 0):
padding = "valid"
else:
padding = "same"
#causal
if (params["layout"][-1] == "C"):
data_format = 'channels_last'
else:
data_format = 'channels_first'
dilation_rate = params["dilation"]
activation = None
use_bias = params["use_bias"]
kernel_initializer = get_initializer(network_initializer)
bias_initializer = 'zeros'
kernel_regularizer = None
bias_regularizer = None
activity_regularizer = None
kernel_constraint = None
bias_constraint = None
layer = keras.layers.Conv3D(out_channels,
kernel_size,
strides=strides,
padding=padding,
data_format=data_format,
dilation_rate=dilation_rate,
activation=activation,
use_bias=use_bias,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
kernel_regularizer=kernel_regularizer,
bias_regularizer=bias_regularizer,
activity_regularizer=activity_regularizer,
kernel_constraint=kernel_constraint,
bias_constraint=bias_constraint)
return layer
def custom_model_layer_fully_connected(params, network_initializer):
units = params["units"]
activation = None
use_bias = params["use_bias"]
kernel_initializer = get_initializer(network_initializer)
bias_initializer = 'zeros'
kernel_regularizer = None
bias_regularizer = None
activity_regularizer = None
kernel_constraint = None
bias_constraint = None
layer = keras.layers.Dense(units,
activation=activation,
use_bias=use_bias,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
kernel_regularizer=kernel_regularizer,
bias_regularizer=bias_regularizer,
activity_regularizer=activity_regularizer,
kernel_constraint=kernel_constraint,
bias_constraint=bias_constraint)
return layer
