Besides models from the packages keras, torch and neuralnet it is also possible to pass a self defined model in form of a named list to the converter class. This enables the interpretation of networks from other packages with all available methods provided by the innsight package.
If you want to create a custom model, your list (e.g. model <- list()) needs at least the keys model$input_dim and model$layers. However, there are other optional keys that can be used to name the input features and output nodes or to test the model for correctness. In summary:
$input_dim
An integer vector with the model input dimension without the batch axis, e.g. for a dense layer with 5 input features use c(5) or for a 1D-convolutional layer with signal length 50 and 4 channels use c(4,50).
$input_names (optional)
A list with the names for each input dimension excluding the batch axis, e.g. for a dense layer with 3 input features use list(c("X1", "X2", "X3")) or for a 1D-convolutional layer with signal length 5 and 2 channels use list(c("C1", "C2"), c("L1","L2","L3","L4","L5")). By default (NULL), the names are generated.
$output_dim (optional)
An integer vector with the model output dimension analogous to $input_dim. This value does not need to be specified and will be calculated otherwise. However, if it is set, the calculated value will be compared with it to avoid errors during converting the model.
$output_names (optional)
A list with the names for each output dimension analogous to $input_names. By default (NULL), the names are generated.
$layers
(see next section)
Example:
The component $layers is again a list that describes each individual layer of the model as another list in correct order. Additionally, depending on the layer type, other keys are required to create a well-defined model. Currently, the following layers are accepted:
$type: 'Dense'
$weight: The weight matrix of the dense layer with shape (dim_out, dim_in).
$bias: The bias vector of the dense layer with length dim_out.
activation_name: The name of the activation function for this dense layer, e.g. 'relu', 'tanh' or 'softmax'.
dim_in (optional): The input dimension of this layer excluding the batch axis e.g. for 3 input features use c(3). This value is not necessary, but helpful to check the format of the weight matrix and the overall correctness of the converted model.
dim_out (optional): The output dimension of this layer excluding the batch axis. This value is not necessary, but helpful to check the format of the weight matrix and the overall correctness of the converted model.
Example:
$type: 'Conv1D' or 'Conv2D'
$weight: The weight array of the convolutional layer with shape (out_channels, in_channels, kernel_length) for 1D or (out_channels, in_channels, kernel_height, kernel_width) for 2D.
$bias: The bias vector of the layer with length out_channels.
$activation_name: The name of the activation function for this layer, e.g. 'relu', 'tanh' or 'softmax'.
$dim_in (optional): The input dimension of this layer according to the format (in_channels, in_length) for 1D or (in_channels, in_height, in_width) for 2D. This value is not necessary, but helpful to check the format of the weight array and the overall correctness of the converted model.
$dim_out (optional): The output dimension of this layer according to the format (out_channels, out_length) for 1D or (out_channels, out_height, out_width) for 2D. This value is not necessary, but helpful to check the format of the weight array and the overall correctness of the converted model.
$stride (optional): The stride of the convolution (single integer for 1D and tuple of two integers for 2D). If this value is not specified, the default values (1D: 1 and 2D: c(1,1)) are used.
$padding (optional): Zero-padding added to the sides of the input before convolution. For 1D-convolution a tuple of the form (pad_left, pad_right) and for 2D-convolution (pad_left, pad_right, pad_top, pad_bottom) is required. If this value is not specified, the default values (1D: c(0,0) and 2D: c(0,0,0,0)) are used.
$dilation (optional): Spacing between kernel elements (single integer for 1D and tuple of two integers for 2D). If this value is not specified, the default values (1D: 1 and 2D: c(1,1)) are used.
Examples:
Create 1D-convolutional layer with: 'in_channels' = 3, 'out_channels' = 8, 'kernel_length' = 2, 'activation_name' = 'tanh' and default values for 'padding', 'stride' and 'dilation':
conv_1D <- list(
type = "Conv1D",
weight = array(rnorm(8*3*2), dim = c(8,3,2)),
bias = rnorm(8),
activation_name = "tanh",
dim_in = c(3, 10), # optional
dim_out = c(8, 9) # optional
)
# Adding layer to model
model$layers$custom_conv1D <- conv_1DCreate 2D-convolutional layer with: 'in_channels' = 3, 'out_channels' = 8, 'kernel_height' = 2, 'kernel_width' = 4, 'activation_name' = 'relu', 'padding' = c(1,1,0,0) and 'dilation' = c(1,2):
$type: 'MaxPooling1D', 'MaxPooling2D', 'AveragePooling1D' or 'AveragePooling2D'
$kernel_size: The size of the pooling window as an integer value for 1D-pooling and an tuple of two integers for 2D-pooling.
$strides (optional): The stride of the pooling window (single integer for 1D and tuple of two integers for 2D). If this value is not specified (NULL), the value of kernel_size will be used.
dim_in (optional): The input dimension of this layer according to the format (in_channels, in_length) for 1D or (in_channels, in_height, in_width) for 2D. This value is not necessary, but helpful to check overall correctness of the converted model.
$dim_out (optional): The output dimension of this layer according to the format (out_channels, out_length) for 1D or (out_channels, out_height, out_width) for 2D. This value is not necessary, but helpful to check overall correctness of the converted model.
Example: Create 2D average pooling with kernel size 'c(2,2)':
$type: 'Flatten'
$dim_in (optional): The input dimension of this layer without the batch dimension.
$dim_out (optional): The output dimension of this layer without the batch dimension.
Example:
flatten <- list(
type = "Flatten"
)
# Adding layer to model
model$layers$custom_flatten <- flatten
library(innsight)
model <- list()
model$input_dim <- 5
model$input_names <- list(c("Feat1", "Feat2", "Feat3", "Feat4", "Feat5"))
model$output_dim <- 2
model$output_names <- list(c("Cat", "no-Cat"))
model$layers$Layer_1 <-
list(
type = "Dense",
weight = matrix(rnorm(5 * 20), 20, 5),
bias = rnorm(20),
activation_name = "tanh",
dim_in = 5L,
dim_out = 20L
)
model$layers$Layer_2 <-
list(
type = "Dense",
weight = matrix(rnorm(20 * 2), 2, 20),
bias = rnorm(2),
activation_name = "softmax",
dim_in = 20L,
dim_out = 2L
)
# Convert the model
converter <- Converter$new(model)
library(innsight)
model <- list()
model$input_dim <- c(3, 30, 30)
model$output_dim <- c(2)
model$layers$Layer_1 <- list(
type = "Conv2D",
weight = array(rnorm(10*3*5*5), dim = c(10,3,5,5)),
bias = rnorm(10),
activation_name = "relu"
)
model$layers$Layer_2 <- list(
type = 'AveragePooling2D',
kernel_size = c(2,2),
dim_in = c(10, 26, 26) # optional
)
model$layers$Layer_3 <- list(
type = 'Conv2D',
weight = array(rnorm(8*10*4*4), dim = c(8,10,4,4)),
bias = rnorm(8),
activation_name = "relu",
padding = c(2,2,3,3),
dim_out = c(8, 16, 14) # optional
)
model$layers$Layer_4 <- list(
type = "AveragePooling2D",
kernel_size = c(2,2),
strides = c(2,3)
)
model$layers$Layer_5 <- list(
type = "Flatten",
dim_out = c(320) # optional
)
model$layers$Layer_6 <- list(
type = "Dense",
weight = array(rnorm(320*2), dim = c(2, 320)),
bias = rnorm(2),
activation_name = "softmax"
)
# Convert the model
converter <- Converter$new(model)