exploration.Rmd

---
title: "Exploratory data analysis"
author: "Guillem Hurault"
date: "`r format(Sys.time(), '%d %B %Y')`"
output:
  html_document:
    toc: true
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE,
                      message = FALSE,
                      warning = FALSE,
                      fig.height = 5,
                      fig.width = 12,
                      dpi = 200)
```

```{r initialisation}
library(tidyverse)

train <- targets::tar_read(training_data)
```

```{r exploration1}
head(train)

summary(train)

pairs(train[, 1:10])
```

## Visualising missing values

```{r exploration-missing}
train %>%
  select(-Y) %>%
  is.na() %>%
  as_tibble() %>%
  mutate(Index = 1:n()) %>%
  pivot_longer(-Index) %>%
  ggplot(aes(x = name, y = Index, fill = value)) +
  geom_tile() +
  labs(x = "Variable", fill = "Missing") +
  coord_cartesian(expand = FALSE) +
  scale_fill_manual(values = c("#000000", "#E69F00")) +
  theme_classic(base_size = 15)
```

## Multicollinearity

For plotting, we impute missing values by 0 (mean of each feature).

```{r exploration-correlation}
train %>%
  select(-Y, -Xb1, -Xb2) %>%
  replace(is.na(.), 0) %>%
  cor() %>%
  corrplot::corrplot.mixed()
```

## Distribution of the outcome variable

```{r exploration-outcome}
table(train[["Y"]]) %>%
  barplot(main = "Count of outcome variable")
```