halflearned

# Native Python packages
import numpy as np
import pandas as pd

# R2pi package
from rpy2.robjects import numpy2ri, pandas2ri
from rpy2.robjects.packages import importr

# R imports (Assuming AER and ivpack are installed in R)
base = importr("base")

halflearned

# Reference: https://github.com/swager/grf/issues/247
# Here I shut off nonlinearity in true model, just for diagnostics

library(tidyverse)
library(grf)

simple_oracle_data <- function(nsamp, 
                               beta=c(2,-1,0,0,0), 
                               te=1,
                               hete=1, 

halflearned

library(tidyverse)

test_fun <- function(n=200, p=20, num.trees=NULL, mtry=NULL) {

  # Creating regressors: x1, ..., xp
  X <- as.data.frame(matrix(rnorm(n*p), n, p))
  colnames(X) <- paste("x", seq_len(p), sep="")

  # Let x1 be the only relevant covariate
  Y <- X[,1]

halflearned

library(grf)

n <- 100

# X ~ Unif[-5,-2]+[2, 5]
X <- matrix(c(seq(-5, -2, length.out=n/2), seq(2, 5, length.out=n/2)), n, 1)

# P( W=1 | X > 0) = 1  and  P( W=1 | X < 0) = 1
W <- matrix(X[,1] > 0, n, 1)

halflearned

# PART II
library(DiceKriging)

n <- 100
k <- 3

# Design (fit.draws in tune_*_forest)
design <- matrix(runif(n*k), n, k)

# If both 'response' and 'noise.var' are constants,

halflearned

library(grf)

n <- 100
k <- 3


X <- matrix(rnorm(n*k), n, k)

# Outcome is constant!
Y_constant <- matrix(rep(0, n))

halflearned

#include <iostream>
#include <set>
#include <vector>


void fill_skipping(std::vector<size_t>& result,
    std::set<size_t>& skip) {

  auto p_skip = skip.begin();
  auto p_result = result.begin();

halflearned

library(grf)

# Some forest
n <- 1000
k <- 3
X <- matrix(runif(n*k), nrow=n, ncol=k)
Y <- matrix(runif(n), nrow=n, ncol=1)
forest <- grf::regression_forest(X, Y)

# Get tree

halflearned

library(grf)

p = 3
n = 2000
sigma = 0.1

X = matrix(2 * runif(n * p) - 1, n, p)
W = rbinom(n, 1, 0.1)
TAU = (X[,1] > 0)
Y = TAU * (W - 1/2) + sigma * rnorm(n)

halflearned

library(grf)

set.seed(123)

# Very simple scenario, but already adversarial
n <- 200
p <- 2
X <- matrix(rnorm(n * p), n, p)
W <- rbinom(p=0.5, size=1, n=n)
eta <-  1