10 library(predictionError)
11 options(amelia.parallel="no",
15 source("irr_dv_simulation_base.R")
18 ## one way to do it is by adding correlation to x.obs and y that isn't in w.
19 ## in other words, the model is missing an important feature of x.obs that's related to y.
20 simulate_data <- function(N, m, B0, Bxy, Bzy, seed, prediction_accuracy=0.73, coder_accuracy=0.8){
23 # make w and y dependent
24 z <- rbinom(N, 1, 0.5)
25 x <- rbinom(N, 1, 0.5)
27 ystar <- Bzy * z + Bxy * x + B0
28 y <- rbinom(N,1,plogis(ystar))
30 # glm(y ~ x + z, family="binomial")
32 df <- data.table(x=x,y=y,ystar=ystar,z=z)
34 df <- df[sample(nrow(df), m), y.obs := y]
36 coder.0.correct <- rbinom(m, 1, coder_accuracy)
37 coder.1.correct <- rbinom(m, 1, coder_accuracy)
39 df[!is.na(y.obs),y.obs.0 := as.numeric((.SD$y.obs & coder.0.correct) | (!.SD$y.obs & !coder.0.correct))]
40 df[!is.na(y.obs),y.obs.1 := as.numeric((.SD$y.obs & coder.1.correct) | (!.SD$y.obs & !coder.1.correct))]
42 odds.y1 <- qlogis(prediction_accuracy)
43 odds.y0 <- qlogis(prediction_accuracy,lower.tail=F)
45 df[y==0,w:=plogis(rlogis(.N,odds.y0))]
46 df[y==1,w:=plogis(rlogis(.N,odds.y1))]
48 df[,w_pred := as.integer(w > 0.5)]
50 print(mean(df$y == df$y.obs.0,na.rm=T))
51 print(mean(df$y == df$y.obs.1,na.rm=T))
53 print(mean(df[x==0]$y == df[x==0]$w_pred))
54 print(mean(df[x==1]$y == df[x==1]$w_pred))
55 print(mean(df$w_pred == df$y))
59 parser <- arg_parser("Simulate data and fit corrected models")
60 parser <- add_argument(parser, "--N", default=5000, help="number of observations of w")
61 parser <- add_argument(parser, "--m", default=500, help="m the number of ground truth observations")
62 parser <- add_argument(parser, "--seed", default=16, help='seed for the rng')
63 parser <- add_argument(parser, "--outfile", help='output file', default='example_2.feather')
64 parser <- add_argument(parser, "--y_explained_variance", help='what proportion of the variance of y can be explained?', default=0.1)
65 parser <- add_argument(parser, "--prediction_accuracy", help='how accurate is the predictive model?', default=0.73)
66 ## parser <- add_argument(parser, "--x_bias_y1", help='how is the classifier biased when y = 1?', default=-0.75)
67 ## parser <- add_argument(parser, "--x_bias_y0", help='how is the classifier biased when y = 0 ?', default=0.75)
68 parser <- add_argument(parser, "--Bxy", help='coefficient of x on y', default=0.3)
69 parser <- add_argument(parser, "--Bzy", help='coeffficient of z on y', default=-0.3)
70 parser <- add_argument(parser, "--outcome_formula", help='formula for the outcome variable', default="y~x+z")
71 parser <- add_argument(parser, "--proxy_formula", help='formula for the proxy variable', default="w_pred~y+y.obs.1+y.obs.0")
72 parser <- add_argument(parser, "--coder_accuracy", help='How accurate are the coders?', default=0.8)
74 args <- parse_args(parser)
81 df <- simulate_data(args$N, args$m, B0, Bxy, Bzy, args$seed, args$prediction_accuracy, args$coder_accuracy)
83 # result <- list('N'=args$N,'m'=args$m,'B0'=B0,'Bxy'=Bxy,'Bzy'=Bzy, 'seed'=args$seed, 'y_explained_variance'=args$y_explained_variance, 'prediction_accuracy'=args$prediction_accuracy, 'x_bias_y0'=args$x_bias_y0,'x_bias_y1'=args$x_bias_y1,'outcome_formula' = args$outcome_formula, 'proxy_formula' = args$proxy_formula)
84 result <- list('N'=args$N,'m'=args$m,'B0'=B0,'Bxy'=Bxy,'Bzy'=Bzy, 'seed'=args$seed, 'y_explained_variance'=args$y_explained_variance, 'prediction_accuracy'=args$prediction_accuracy, 'outcome_formula' = args$outcome_formula, 'proxy_formula' = args$proxy_formula)
86 outline <- run_simulation_depvar(df, result, outcome_formula = as.formula(args$outcome_formula), proxy_formula = as.formula(args$proxy_formula))
88 outfile_lock <- lock(paste0(args$outfile, '_lock'),exclusive=TRUE)
90 if(file.exists(args$outfile)){
91 logdata <- read_feather(args$outfile)
92 logdata <- rbind(logdata,as.data.table(outline),fill=TRUE)
94 logdata <- as.data.table(outline)
98 write_feather(logdata, args$outfile)