forked from child-growth/ki-growth-prediction
-
Notifications
You must be signed in to change notification settings - Fork 0
/
reproducible_example.R
176 lines (126 loc) · 5.07 KB
/
reproducible_example.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
rm(list=ls())
source(paste0(here::here(), "/0-config.R"))
source(paste0(here::here(), "/functions/0_calc_empirical_IC.R"))
library(hal9001)
library(washb)
#Load data
d <- readRDS(paste0(data_dir,"analysis_data.RDS")) %>% filter(studyid=="MAL-ED", country=="INDIA")
#Subset to control arms
#d <- d %>% filter(arm=="" | arm=="Control" | arm=="Iron and folic acid supplementation" | arm=="Iron Folic Acid")
y_name = "whz_24"
x_names = c("sex", "W_mage", "W_meducyrs", "W_nrooms", "W_mwtkg","W_mhtcm","W_mbmi",
"miss_W_mage","miss_W_meducyrs", "miss_W_nrooms", "miss_W_mwtkg", "miss_W_mhtcm", "miss_W_mbmi" , "arm", "waz_birth")
dfull <- d %>% ungroup() %>% select(all_of(c("studyid","country",y_name, x_names)))
#temp: complete cases
dim(dfull)
table(!is.na(dfull$waz_birth))
dfull <- dfull[complete.cases(dfull),]
dfull <- droplevels(dfull)
dim(dfull)
n=nrow(dfull)
Y <- as.numeric(as.matrix(dfull %>% select(all_of(y_name))))
X <- dfull %>% ungroup() %>%
select(all_of(x_names)) %>%
mutate_if(sapply(., is.character), as.factor)
NZV <- nearZeroVar(X, saveMetrics = TRUE)
X <- X[,!NZV$zeroVar]
X <-design_matrix(as.data.frame(X))
# hal_fit_list <- readRDS(paste0(model_dir,"res_wazBirth_whz24_hal.RDS"))
#
# hal_fit<-hal_fit_list$`MAL-ED-INDIA`
fit <- fit_hal(X = X,
Y = Y,
smoothness_orders = 0,
return_x_basis = TRUE,
family = "gaussian",
num_knots = hal9001:::num_knots_generator(
max_degree = ifelse(ncol(X) >= 20, 2, 3),
smoothness_orders = 0,
base_num_knots_0 = max(100, ceiling(sqrt(n)))
))
fit$coefs[fit$coefs!=0]
#t-statistic based mean outcome
library(distributions3)
n <- length(Y)
T_stat <- StudentsT(df = n-1)
mean(Y) + quantile(T_stat, 0.12 / 2) * sd(Y) / sqrt(n)
mean(Y) + quantile(T_stat, 1 - 0.12 / 2) * sd(Y) / sqrt(n)
#washb mean + robust CI
res_mean <- data.frame(washb::washb_mean(Y=Y, id=1:length(Y), print=FALSE))
res_mean
#Get HAL predictions - just with the data used to fit the model
#res_hal <- get_pred_empirical_ci(fit=hal_fit_list[[i]]$fit_init, X=hal_fit_list[[i]]$X, Y=hal_fit_list[[i]]$Y, n=hal_fit_list[[i]]$n))
Y_hat <- predict(fit, new_data = X)
init_coef <- fit$coefs[-1]
nonzero_col <- which(init_coef != 0)
init_coef_nonzero <- init_coef[nonzero_col]
basis_mat <- as.matrix(fit$x_basis)
basis_mat <- as.matrix(basis_mat[, nonzero_col])
#cal_IC_for_beta_cont <- function(X, Y, Y_hat, beta_n){
# 1. calculate score: X'(Y-Y_hat)
score <- sweep(basis_mat, 1, (Y - Y_hat), `*`)
# 2. calculate E_{P_n}(X'X)^(-1)
d_scaler = solve(t(basis_mat) %*% basis_mat)
# 3. calculate influence curves
IC_beta <- t(d_scaler %*% t(score))
# return(IC)
# }
#IC_phi <- cal_IC_for_phi(X_new = basis_mat, beta_n = init_coef_nonzero, IC_beta)
beta_n = init_coef_nonzero
X_new=basis_mat
d_phi_scaler_new <- as.vector(exp(- beta_n %*% t(X_new)) / ((1 + exp(- beta_n %*% t(X_new)))^2))
d_phi_new <- sweep(X_new, 1, d_phi_scaler_new, `*`)
IC_phi = diag(d_phi_new %*% t(IC_beta))
se_IC <- sqrt(var(IC_phi)/n)
res <- data.frame(predY = mean(Y_hat))
res$se <- se_IC
res$lb <- res$predY - 1.96 * res$se
res$ub <- res$predY + 1.96 * res$se
res
# get_counterfactual_empirical_ci
# function(fit, X, Y, n, IC_beta){
#
# if(!is.null(fit)){
# Y_hat_init <- predict(fit, new_data = X)
#
# init_coef <- fit$coefs[-1]
# nonzero_col <- which(init_coef != 0)
# init_coef_nonzero <- init_coef[nonzero_col]
#
# x_basis <- make_design_matrix(as.matrix(X), fit$basis_list, p_reserve = 0.75)
# x_basis <- as.matrix(x_basis[, nonzero_col])
#
# IC_phi <- NULL
# try(IC_phi <- cal_IC_for_phi(X_new = x_basis, beta_n = init_coef_nonzero, IC_beta))
#
# if(!is.null(IC_phi)){
# se_IC <- sqrt(var(IC_phi)/n)
#
# res <- data.frame(predY = mean(Y_hat_init))
# res$se <- se_IC
# res$lb <- res$predY - 1.96 * res$se
# res$ub <- res$predY + 1.96 * res$se
#
# return(res)
# }else{
# return(NULL)
# }
# }
#
# }
X1 <- X %>% mutate(waz_birth=waz_birth+0.5)
# function(fit, X, Y, n, IC_beta){
Y_hat_init <- predict(fit, new_data = X1)
init_coef <- fit$coefs[-1]
nonzero_col <- which(init_coef != 0)
init_coef_nonzero <- init_coef[nonzero_col]
x_basis1 <- make_design_matrix(as.matrix(X1), fit$basis_list, p_reserve = 0.75)
x_basis1 <- as.matrix(x_basis1[, nonzero_col])
IC_phi <- cal_IC_for_phi(X_new = x_basis1, beta_n = init_coef_nonzero, IC_beta)
se_IC <- sqrt(var(IC_phi)/n)
res_counterfactual <- data.frame(predY = mean(Y_hat_init))
res_counterfactual$se <- se_IC
res_counterfactual$lb <- res_counterfactual$predY - 1.96 * res_counterfactual$se
res_counterfactual$ub <- res_counterfactual$predY + 1.96 * res_counterfactual$se
res
res_counterfactual