deeplearning (version 0.1.0)

print_weight: Prints out the weight of a deep neural network

Description

This function prints out the weight in a heat map, 3D surface, or histogram

Usage

print_weight(darch, num_of_layer, show_derivative = F, type = "heatmap")

Arguments

darch

DArch instance

num_of_layer

the number of the layer to print

show_derivative

T to show the weight value. F to show the percentage weight change in the finetuning stage. This helps spot the network saturation problem.

type

type of the graph. It supports "heatmap", "surface", and "histogram"

Examples

# Example of Regression

input <- matrix(runif(1000), 500, 2)
input_valid <- matrix(runif(100), 50, 2)
target <- rowSums(input + input^2)
target_valid <- rowSums(input_valid + input_valid^2)
# create a new deep neural network for classificaiton
dnn_regression <- new_dnn(
 c(2, 50, 50, 20, 1),  # The layer structure of the deep neural network.
 # The first element is the number of input variables.
 # The last element is the number of output variables.
 hidden_layer_default = rectified_linear_unit_function,
 # for hidden layers, use rectified_linear_unit_function
 output_layer_default = linearUnitDerivative
 # for regression, use linearUnitDerivative function
)

# print the layer weights
# this function can print heatmap, histogram, or a surface
print_weight(dnn_regression, 1, type = "heatmap")

print_weight(dnn_regression, 2, type = "surface")

print_weight(dnn_regression, 3, type = "histogram")