Update README.md

Author: momenulhaque

Diff for: README.md

@@ -3,41 +3,74 @@
 ## How to install and apply DC-AIPW and DC-TMLE on a data set
 
 ```{r}
-credentials::set_github_pat() # it asks for token. 
 devtools::install_github("momenulhaque/Crossfit") # it will install the package
-library(Crossfit) # Now the package is ready to use
+library(Crossfit) 
+```
+Now the package is ready to use. It supports applying both DC-AIPW and DC-TMLE for two cases-
+
+### Case 1: Without parallelization
 
+ 1. Install the required R packages
 
-################# Without parallelization ##################
+```{r}
 library(tidyverse)
 require(furrr)
 require(tibble)
 require(SuperLearner)
+```
+  2. Define the learners that you want to use in superlearner training
 
-# I used only two learners
+```{r}
 #Logistic Regression
+  SL.glm.DCDR <- function(...){
+    SL.glm(...)
+  }
+  
+#4 degree GAM
+  SL.gam4.DCDR <- function(...){
+    SL.gam(..., deg.gam=4)
+  }
+  
+#6 degree GAM
+  SL.gam6.DCDR <- function(...){
+    SL.gam(..., deg.gam=6)
+  }
+  
+#Neural Network
+  SL.nnet.DCDR <- function(...){
+    SL.nnet(..., size=4)
+  }
+  
+#Random forest
+  SL.randomForest.DCDR <- function(...){
+    SL.randomForest(..., ntree=500, nodesize=20)
+  }
 
-SL.glm.DCDR <- function(...){
-  SL.glm(...)
-}
+#Empirical mean
+  SL.mean.DCDR <- function(...){
+    SL.mean(...)
+  }
+  
 
-#4 degree GAM
-SL.gam4.DCDR <- function(...){
-  SL.gam(..., deg.gam=4)
-}
+learners <- c("SL.glm.DCDR", "SL.gam4.DCDR", "SL.gam6.DCDR", "SL.nnet.DCDR", "SL.randomForest.DCDR", "SL.mean.DCDR")
+```
 
-# data can be from Crossfit package
+  3. Defining the data you want to use
+ 
+```{r}
+# Read the data set that you want to use. An example data set "data" can be found in this package.
 df = data 
+```
+  3. Defining model parameters
 
+```{r}
 exposure="statin"
 outcome="Y"
 
-
 covarsT <- c("age", "ldl_log", "risk_score") # covariate for exposure
 covarsO <- c("age", "ldl_log", "risk_score") # covariate for outcome
 
-learners <- c("SL.glm.DCDR", "SL.gam4.DCDR")
-
+# Here `V=5' indicates the number of cross-validation folds that is applied in the superlearner training.
 control <- SuperLearner.CV.control(V=5)
 
 ## Wrapper functions
@@ -53,14 +86,21 @@ tmle_sim <- function(df, num_cf = 3, n_split, seed){
                                  num_cf, n_split, rand_split = TRUE, seed)
   return(tmle_output)
 }
-
-
-
+```
+  3. Estimating the average causal effect
+  Here the parameters **num_cf** is the number of repeatation, **n_split** is the number of splits and **seed** can be usefull for comparing different methods.
+```{r}
 aipw_result_p3 = aipw_sim(df=data, num_cf = 3, n_split = 3, seed = 123)
 tmle_result_p3 = tmle_sim(df=data, num_cf = 3, n_split = 3, seed = 123)
+```
+
+### Case 2: With parallelization
+The parallelization is very usefull while simulation study is conducted for large number of times. The steps are similar to case 1, except some additional steps-
 
+ 1. Do the steps 1-3 that is described in case 1 under parallel packages. 
 
 
+```{r}
 ############### With parallelization ##############
 library(parallel)
 cl <- makeCluster(detectCores())
@@ -73,61 +113,24 @@ parallel::clusterEvalQ(cl, {
   require(tibble)
   require(SuperLearner)
   
- #Logistic Regression
-  
-  SL.glm.DCDR <- function(...){
-    SL.glm(...)
-  }
-  
-  #4 degree GAM
-  SL.gam4.DCDR <- function(...){
-    SL.gam(..., deg.gam=4)
-  }
-  
-  df = data # data is from Crossfit package
-  
-  exposure="statin"
-  outcome="Y"
-  
-  
-  covarsT <- c("age", "ldl_log", "risk_score") # covariate for exposure
-  covarsO <- c("age", "ldl_log", "risk_score") # covariate for outcome
-  
-  learners <- c("SL.glm.DCDR", "SL.gam4.DCDR")
-  
-  control <- SuperLearner.CV.control(V=5)
-  
-  
-  ## Wrapper functions
-  
-  aipw_sim <- function(df, num_cf = 3, n_split, seed){
-    aipw_output <- aipw_multiple_p(df, exposure, outcome, covarsT, covarsO, learners, control,
-                                   num_cf, n_split, rand_split = TRUE, seed)
-    return(aipw_output)
-  }
-  
-  tmle_sim <- function(df, num_cf = 3, n_split, seed){
-    tmle_output <- tmle_multiple_p(df, exposure, outcome, covarsT, covarsO, learners, control,
-                                   num_cf, n_split, rand_split = TRUE, seed)
-    return(tmle_output)
-  }
-
+ # Run the codes from Step 1
+ # Run the codes from Step 2
+ # Run the codes from Step 3
 
 })
 
+```
 
 
-# I just made a list of two same `data'  to apply clusterMap function.
-
+ 2. Apply the following codes for estimating average causal effect under parallel packages-
+ All the data set should be stored in a list object. I just made a list of two data sets using the same data `data'  to apply clusterMap function.
 
+```{r}
 aipw_result_p3 = clusterMap(cl, function(df, seed) aipw_sim(df, num_cf = 3, n_split = 3, seed),
                             df=list(data, data) , seed=list(1:2))
 
-
 tmle_result_p3 = clusterMap(cl, function(df, seed) tmle_sim(df, num_cf = 3, n_split = 3, seed),
                             df=list(data, data) , seed=list(1:2))
 
 
-
-
 ```