1-explore.Rmd

---
title: "1-explore"
author: "bernard-liew"
date: "2021-03-29"
output: workflowr::wflow_html
editor_options:
  chunk_output_type: console
---

# Package

```{r, include = FALSE}
knitr::opts_chunk$set(eval = FALSE)
```


```{r}
# Helper
library (tidyverse)
library (skimr)
library (DataExplorer)
library (janitor)
library (rsample)

# Import
library (rio)

# Missing
library (VIM)
library (naniar)
library (mice)
library (NADIA)
```

Model for Improvement of neckpain: filter by VAS_neckpain_0 > 1.5 AND 0 missing values for all the obseravtions who entered into the full model.

Model for Improvement of armpain: filter by VAS_armpain_0 > 1.5  AND 0 missing values for all the obseravtions who entered into the full model. 

Model for Improvement of disability: filter by disability_0 >= 7  AND 0 missing values for all the obseravtions who entered into the full model. 


# Import

```{r}
df <- import ("data/neck_pain_database_3001.xlsx", sheet = "DATA")
keys <- import ("data/neck_pain_database_3001.xlsx", sheet = "LEGEND")
```

# Tidy

## Convert factors

```{r}
var_as_factors <- keys[["Type of variable"]] == "Factor"

df <- df %>%
  mutate_if (var_as_factors, ~.x %>% 
            as.character() %>% as.factor()) %>% 
  rename (imp_np = improvem_NECKpain,
          imp_ap = improvem_ARMpain,
          imp_dis = improvem_DISAB)

```


## Explore data

```{r}
skim (df)

table (df$improvem_NECKpain, df$improvem_ARMpain, df$improvem_DISAB)

```

```{r}
visdat::vis_dat(df, sort_type = TRUE)
visdat::vis_miss(df, cluster = TRUE)
```


## Missingness

```{r}
aggr(df)

barMiss(df %>%
             select (imp_np, imp_dis))

spineMiss(df %>%
             select (imp_np, imp_dis))

spineMiss(df %>%
             select (imp_np, imp_ap))
```

```{r}
np <- df %>% 
  select (-c(imp_ap, imp_dis)) %>%
  purrr::discard(~sum(is.na(.x))/length(.x)* 100 >= 50) %>%
  rename (outcome = imp_np)

ap <- df %>% 
  select (-c(imp_np, imp_dis)) %>%
  purrr::discard(~sum(is.na(.x))/length(.x)* 100 >= 50)%>%
  rename (outcome = imp_ap)

dis <- df %>% 
  select (-c(imp_np, imp_ap)) %>%
  purrr::discard(~sum(is.na(.x))/length(.x)* 100 >= 60)%>%
  rename (outcome = imp_dis)

df_list <- list (np = vector ("list"),
                 ap = vector ("list"),
                 dis = vector ("list"))

df_list$np$orig <- np
df_list$ap$orig <- ap
df_list$dis$orig <- dis
```



```{r}

for (n in seq_along (df_list)) {
  set.seed(123)
  
  df_split <- initial_split(df_list[[n]]$orig, prop = 0.80, strata = outcome)
  train <- training(df_split)
  test  <-  testing(df_split)
  
  df_list[[n]]$train <- train
  df_list[[n]]$test <- test
}



# Save the split information for an 80/20 split of the data


```

```{r}

for (n in seq_along (df_list)) {
  
  train_impute <- mice (df_list[[n]]$train)
  df_list[[n]]$train_imp <- complete (train_impute)
  
  test_impute <- mice.reuse(train_impute,
                           df_list[[n]]$test)

  df_list[[n]]$test_imp <- test_impute[[1]]
}

# train_impute <- mice (train)
# 
# train2 <- complete (train_impute)
# test_impute <- mice.reuse(train_impute,
#                           test)
# test2 <- test_impute[[1]]
# 
dat <- list (data = df,
             df_list = df_list)
```


## Save 

```{r}
saveRDS(dat,
        "output/df.RDS")
```