CausalStump: Fit a Gaussian process or Student-t process to the treatment...

In mazphilip/CausalStump: CausalStump: A structural Gaussian process implementation for Causal Inference

Description

Fit a Gaussian process or Student-t process to the treatment and control surface.

Usage

CausalStump(y, X, z, w, pscore, kernelfun = "SE", myoptim = "Nadam",
  maxiter = 5000, tol = 1e-04, prior = FALSE, nu = 200,
  nsampling = 5000, learning_rate = 0.01, beta1 = 0.9, beta2 = 0.999,
  momentum = 0)

Arguments

y	A vector
X	A data.frame
z	A vector
w	A vector (optional)
pscore	A vector (optional)
kernelfun	A string (default: SE)
myoptim	A string (default: Gradient Descent – GD)
maxiter	(default: 5000)
tol	(default: 1e-4)
prior	A logic statement (default: FALSE)
nu	A value (default: 200)
nsampling	A value (default: 5000) number of samples for prediction
learning_rate	(default: 0.01)
beta1	(default: 0.9)
beta2	(default: 0.999)
momentum	(default: 0.0)

Value

The function returns the fitted process as a CausalStump class object

Examples

#Generate data
n = 120
Z = rbinom(n, 1, 0.3)
X0 = runif(n-sum(Z), min = 20, max = 40)
X = data.frame(matrix(NaN,n,1))
X[Z==1,] = X1; X[Z==0,] = X0
y0_true = as.matrix(72 + 3 * sqrt(X))
y1_true = as.matrix(90 + exp(0.06 * X))
Y0 = rnorm(n, mean = y0_true, sd = 1)
Y1 = rnorm(n, mean = y1_true, sd = 1)
Y = Y0*(1-Z) + Y1*Z
mystump <- CausalStump(Y,X,Z)

mazphilip/CausalStump documentation built on May 31, 2019, 3:51 a.m.