nn_helper: Helper function for neural networks fitted by neuralnet

In postDoubleR: Post Double Selection with Double Machine Learning

Description

Helper function that dispatches to neuralnet for the double ML estimation (see details).

Usage

nn_helper(X, Y, W, neural.net.Y = NULL, neural.net.W = NULL,
  standardize = TRUE, standardization.method = "min-max")

Arguments

X	A matrix of covariates (must be all numeric)
Y	A vector of the target variable, of same length as the number of rows of Y, must be numeric
W	A vector of the treatment variable, of same length as the number of rows of X, must be numeric
neural.net.Y	A model specification for Y, see neuralnet
neural.net.W	A model specification for W, see neuralnet
standardize	Whether to standardize the data before starting the computation, defaults to TRUE.
standardization.method	How to standardize data, defaults to min-max, also offers "Z-transform", "Unit-Scale" and "Mean-Scale"

Details

For a more steamlined usage, default arguments as implemented in the neuralnet package are passed to both networks during fitting, unless specified otherwise. Also, any attempt to set the formula or data arguments of neuralnet will be ignored and rewritted with internal structures. The function will print a warning if this happens.

Value

A list with two elements: The fitted W model and the fitted Y model.

Examples

  n = 2000; p = 3
  X = matrix(rnorm(n*p), n, p)
  W = rbinom(n, 1, 0.4 + 0.2 * (X[,1] > 0))
  Y = rbinom(n, 1, 0.2 + 0.2 * (X[,2] > 0) + W * 0.1)

# note that this neural network can fail to converge

 nn_helper( X,
            Y,
            W,
            neural.net.W = list( act.fct = "logistic" ),
            neural.net.Y = list( act.fc = "logistic" ))

postDoubleR documentation built on Oct. 7, 2019, 5:05 p.m.