---
title: "Imputations"
output:
    rmarkdown::html_vignette:
        toc: true
description: >
  Incorporate imputation approach as another source of variability in the generated space of cluster solutions.
vignette: >
  %\VignetteIndexEntry{Imputations}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

<style>
div.aside { background-color:#fff2e6; }
</style>

```{r, include = FALSE}
knitr::opts_chunk$set(
    collapse = TRUE,
    comment = "#>"
)
```

Download a copy of the vignette to follow along here: [imputations.Rmd](https://raw.githubusercontent.com/BRANCHlab/metasnf/main/vignettes/imputations.Rmd)

Missing data can be difficult to handle, especially in the context of unsupervised learning.
In a supervised setting, multiply imputed datasets can be used to generate pooled estimates of model coefficients.

A somewhat analogous process is demonstrated in the code below.
Here, we pretend we've generated two different imputations of the data, `data_list_imp1` and `data_list_imp2`.
The mock code below happens to use the base, unimputed data twice for simplicity.

Separate cluster solutions are generated for the two sets of imputed data, which then have their corresponding solutions matrices stacked together and appended with an `imputation` column that indicates which imputed dataset was used to generate that particular cluster solution.

Moving through the rest of the meta clustering pipeline, the influence of the imputation on meta clustering structure or on separation of other variables in the data can be easily visualized in the `adjusted_rand_index_heatmap` function through the use of `ComplexHeatmap` annotations.

```{r eval = FALSE}
library(metasnf)

# First imputed dataset
data_list_imp1 <- generate_data_list(
    list(abcd_subc_v, "subcortical_volume", "neuroimaging", "continuous"),
    list(abcd_h_income, "household_income", "demographics", "continuous"),
    list(abcd_pubertal, "pubertal_status", "demographics", "continuous"),
    list(abcd_anxiety, "anxiety", "behaviour", "ordinal"),
    list(abcd_depress, "depressed", "behaviour", "ordinal"),
    uid = "patient"
)

# Second imputed dataset
data_list_imp2 <- generate_data_list(
    list(abcd_subc_v, "subcortical_volume", "neuroimaging", "continuous"),
    list(abcd_h_income, "household_income", "demographics", "continuous"),
    list(abcd_pubertal, "pubertal_status", "demographics", "continuous"),
    list(abcd_anxiety, "anxiety", "behaviour", "ordinal"),
    list(abcd_depress, "depressed", "behaviour", "ordinal"),
    uid = "patient"
)

settings_matrix <- generate_settings_matrix(
    data_list,
    nrow = 10,
    min_k = 20,
    max_k = 50,
    seed = 42
)

# Generation of 20 cluster solutions
solutions_matrix_imp1 <- batch_snf(data_list_imp1, settings_matrix)
solutions_matrix_imp2 <- batch_snf(data_list_imp2, settings_matrix)

solutions_matrix_imp1$"imputation" <- 1
solutions_matrix_imp1$"imputation" <- 2

# Create a stacked solution matrix that stores solutions from both imputations
solutions_matrix <- rbind(solutions_matrix_imp1, solutions_matrix_imp2)

# Calculate pairwise similarities across all solutions
# (Including across imputations)
solutions_matrix_aris <- calc_aris(solutions_matrix)

meta_cluster_order <- get_matrix_order(solutions_matrix_aris)

# Base heatmap for identifying meta clusters
ari_hm <- adjusted_rand_index_heatmap(
    solutions_matrix_aris,
    order = meta_cluster_order
)

# Identify meta cluster boundaries
shiny_annotator(ari_hm)
split_vec <- c(2, 5, 12, 17)

ari_mc_hm <- adjusted_rand_index_heatmap(
    solutions_matrix_aris,
    order = meta_cluster_order,
    split_vector = split_vec
)

# Calculate how variables are distributed across solutions
extended_solutions_matrix <- extend_solutions(
    solutions_matrix,
    data_list = data_list
)

# Visualize influence of imputation on meta clustering results
annotated_ari_hm <- adjusted_rand_index_heatmap(
    solutions_matrix_aris,
    order = meta_cluster_order,
    split_vector = split_vec,
    data = extended_solutions_matrix,
    top_hm = list(
        "Depression p-value" = "cbcl_depress_r_pval",
        "Anxiety p-value" = "cbcl_anxiety_r_pval"
    ),
    left_hm = list(
        "Imputation" = "imputation"
    ),
    annotation_colours = list(
        "Depression p-value" = colour_scale(
            extended_solutions_matrix$"cbcl_depress_r_pval",
            min_colour = "purple",
            max_colour = "black"
        ),
        "Anxiety p-value" = colour_scale(
            extended_solutions_matrix$"cbcl_anxiety_r_pval",
            min_colour = "green",
            max_colour = "black"
        ),
        "Imputation" = colour_scale(
            extended_solutions_matrix$"mean_pval",
            min_colour = "lightblue",
            max_colour = "black"
        )
    )
)
```