### EXAMPLE 2_b: demonstrates how measurement error can lead to a type sign error in a covariate
### This is the same as example 2, only instead of x->k we have k->x.
### Even when you have a good predictor, if it's biased against a covariate you can get the wrong sign.
### Even when you include the proxy variable in the regression.
### But with some ground truth and multiple imputation, you can fix it.

library(argparser)
library(mecor)
library(ggplot2)
library(data.table)
library(filelock)
library(arrow)
library(Amelia)
library(Zelig)
library(predictionError)
options(amelia.parallel="no",
        amelia.ncpus=1)

source("simulation_base.R")

## SETUP:
### we want to estimate x -> y; x is MAR
### we have x -> k; k -> w; x -> w is used to predict x via the model w.
### A realistic scenario is that we have an NLP model predicting something like "racial harassment" in social media comments
### The labels x are binary, but the model provides a continuous predictor

### simulation:
#### how much power do we get from the model in the first place? (sweeping N and m)
#### 

simulate_data <- function(N, m, B0=0, Bxy=0.2, Bgy=-0.2, Bgx=0.2, y_explained_variance=0.025, gx_explained_variance=0.15, prediction_accuracy=0.73, seed=1){
    set.seed(seed)
    g <- rbinom(N, 1, 0.5)

    x.var.epsilon <- var(Bgx *g) * ((1-gx_explained_variance)/gx_explained_variance)
    x.epsilon <- rnorm(N,sd=sqrt(x.var.epsilon))
    xprime <- Bgx * g + x.epsilon
    x <- as.integer(logistic(scale(xprime)) > 0.5)

    y.var.epsilon <- (var(Bgy * g) + var(Bxy *x) + 2*cov(Bxy*x,Bgy*g)) * ((1-y_explained_variance)/y_explained_variance)
    y.epsilon <- rnorm(N, sd = sqrt(y.var.epsilon))
    y <- Bgy * g + Bxy * x + y.epsilon

    df <- data.table(x=x,xprime=xprime,y=y,g=g)

    if(m < N){
        df <- df[sample(nrow(df), m), x.obs := x]
    } else {
        df <- df[, x.obs := x]
    }

    df <- df[,w_pred:=x]
    df <- df[sample(1:N,(1-prediction_accuracy)*N),w_pred:=(w_pred-1)**2]
    w <- predict(glm(x ~ w_pred,data=df,family=binomial(link='logit')),type='response')
    df <- df[,':='(w=w, w_pred = w_pred)]
    return(df)
}

parser <- arg_parser("Simulate data and fit corrected models")
parser <- add_argument(parser, "--N", default=500, help="number of observations of w")
parser <- add_argument(parser, "--m", default=100, help="m the number of ground truth observations")
parser <- add_argument(parser, "--seed", default=4321, help='seed for the rng')
parser <- add_argument(parser, "--outfile", help='output file', default='example_1.feather')
parser <- add_argument(parser, "--y_explained_variance", help='what proportion of the variance of y can be explained?', default=0.005)
parser <- add_argument(parser, "--gx_explained_variance", help='what proportion of the variance of x can be explained by g?', default=0.15)
parser <- add_argument(parser, "--prediction_accuracy", help='how accurate is the predictive model?', default=0.73)

args <- parse_args(parser)

B0 <- 0
Bxy <- 0.2
Bgy <- -0.2
Bgx <- 0.4

df <- simulate_data(args$N, args$m, B0, Bxy, Bgy, Bgx, seed=args$seed, y_explained_variance = args$y_explained_variance, gx_explained_variance = args$gx_explained_variance, prediction_accuracy=args$prediction_accuracy)

result <- list('N'=args$N,'m'=args$m,'B0'=B0,'Bxy'=Bxy,'Bgy'=Bgy, 'Bgx'=Bgx, 'seed'=args$seed, 'y_explained_variance' = args$y_explained_variance, 'gx_explained_variance' = args$gx_explained_variance, "prediction_accuracy"=args$prediction_accuracy)
outline <- run_simulation(df, result)

outfile_lock <- lock(paste0(args$outfile, '_lock'),exclusive=TRUE)
if(file.exists(args$outfile)){
    logdata <- read_feather(args$outfile)
    logdata <- rbind(logdata,as.data.table(outline))
} else {
    logdata <- as.data.table(outline)
}

print(outline)
write_feather(logdata, args$outfile)
unlock(outfile_lock)