SPOTMisc (version 1.19.52)

mapX2FLAGS: Map x parameters to a list of named values

Description

numerical parameters are mapped to their meanings, e.g., x[1] to "dropout rate".

Usage

mapX2FLAGS(x, model = "dl")

Value

FLAGS named list (parameter names as specified in getModelConf), e.g., for "dl": dropout, dropoutfac, units, unitsfact, learning_rate, epochs, beta_1, beta_2, layers, epsilon, optimizer

Arguments

x

matrix input values.

model

(char) network type, e.g., "cnn" or "dl". Default: "dl"

Details

For a "dl" network, the parameter vector x is mapped to the following FLAGS:

x[1]: dropout

dropout rate first layer.

x[2]: dropoutfac

dropout factor (multiplier).

x[3]: units

number of units in the first layer.

x[4]: unitsfact

units factor (multiplier).

x[5]: learning_rate

learning rate for optimizer. See, e.g.: link{optimizer_sgd}

x[6]: epochs

number of training epochs.

x[7]: beta_1

The exponential decay rate for the 1st moment estimates. float, 0 < beta < 1. Generally close to 1.

x[8]: beta_2

The exponential decay rate for the 2nd moment estimates. float, 0 < beta < 1. Generally close to 1.

x[9]: layers

number of layers.

x[10]: epsilon

float >= 0. Fuzz factor. If NULL, defaults to k_epsilon().

x[11]: optimizer

integer. Specifies optimizer.

Examples

## First example: dense neural net
x <- getModelConf(list(model="dl"))$defaults
mapX2FLAGS(x=x, model = "dl")
## Second example: convnet
x <- getModelConf(list(model="cnn"))$defaults
mapX2FLAGS(x=x, model = "cnn")