Dashboards/ml/global.R

library(shiny)
library(shinyjs)
library(shinydashboard)
#library("wesanderson") # ayyyy lmao hipster af
library(dplyr)
library(ggplot2)
library(grid) # for the info box
library(plotly)
library(shinycssloaders)

# make shiny non-stupid
#options(shiny.launch.browser = FALSE) # i am a big girl and can tie my own laces
#options(shiny.port = 8000) # don't change the port every time
#options(shiny.host = '0.0.0.0') # This means "listen to all addresses on all interfaces"
#options(width=120)
#options(DT.options = list(scrollX = TRUE))

# FIXME: get rid of this hardcoded thing which i'm only reading in to have resampling types ahead of loading the real files
if (interactive()){
  print("Interactive Session, using home directories")
  data_dir = "~/git/"
} else {
  data_dir = "/srv/shiny-server/git/"
}

thing  = read.csv(paste0(data_dir, "Data/ml_combined/genes/pnca_70_30_actual.csv"))

# list of splits
split_type = c(
  "cd_7030", 
  "cd_8020", 
  "cd_sl",
  "none"
)

split_file = c(
  "_70_30_complete", 
  "_80_20_complete", 
  "_sl_complete", 
  "_none_complete"
)

# necessary because the names will be wrong otherwise
split_map = data.frame(
  files=c(
    "_70_30_complete",
    "_80_20_complete",
    "_sl_complete",
    "_none"
  ),
  splits=c(
    "cd_7030",
    "cd_8020",
    "cd_sl",
    "none"
  )
)

metadata_cols = c("n_training_size", "n_test_size", "n_trainingY_ratio", "n_testY_ratio", "resampling", "n_features")

# hardcoded list of drugs
drug = c("ethambutol", "isoniazid", "pyrazinamide", "rifampicin", "streptomycin")
gene = c("embb", "katg", "pnca", "rpob", "gid")
combo = data.frame(drug, gene)

# Loader for per-gene CSVs
#"/home/sethp/git/Data/ml_combined/genes/pnca_70_30_complete.csv"
loaded_files=list()
for (x in gene) {
  #x=tolower(x)
  for (split in split_file){
    filedata = paste0(x, split)
    filename = paste0(data_dir,'LSHTM_ML/output/genes/',x,split,'.csv')
    
    #print(c(filename))
    #load_name=paste0(combo[gene==x,"drug"],'_',split_map['splits'][split_map['files']==split])
    load_name=paste0(x,'_baselineC_',split_map['splits'][split_map['files']==split])
    #print(load_name)
    # try() on its own is fine here because we don't need to do anything if it fails
    try({loaded_files[[load_name]] = read.csv(filename)})
  }
}
# Funky loader for combined data
for (x in gene) {
  for (ac in c('_actual','_complete', '_FS')){
    for (gene_count in c(1:6)){
      load_name=paste0(gene_count, "genes_logo_skf_BT_", x, ac)
      filename = paste0(data_dir,'LSHTM_ML/output/combined/',load_name, ".csv")
      store_name=paste0(gene_count, "genes_logo_skf_BT_", x, ac)
      # tryCatch is necessary here rather than try() because we need to do more
      # manipulation afterwards (throwing away the column after loading)
      load_successful=TRUE
      tryCatch({temp_df = read.csv(filename)},error=function(e){load_successful<<-FALSE})
      if (load_successful){
        temp_df=temp_df[, 2:ncol(temp_df)] # throw away first column
        loaded_files[[store_name]] = temp_df
        print(paste0("loaded file: ", filename, "into var: ", store_name))
      }
    }
  }
}

scores=c("F1", "ROC_AUC", "JCC", "MCC", "Accuracy", "Recall", "Precision")

resample_types <<- unique(thing$resampling)

makeplot = function(x, # the DataFrame to plot
                    selection, # scoring method e.g. 'MCC'
                    resampler, # resampling type e.g. 'none'
                    display_infobox         = TRUE, # display the infobox on top of the plot
                    display_combined        = TRUE, # show stuff that only applies to "combined model" plots
                    gene                    = 'NOT SET', # used only for the info box
                    drug                    = 'NOT SET', # used only for the info box
                    combined_training_genes = '999' # used only for the info box
){
  plot_data    = x[x$resampling==resampler,]
  y_coord_min = min(plot_data[selection])
  
  if (y_coord_min > 0) {
    y_coord_min = 0
  }
  if (display_infobox) {
    metadata=t(plot_data[1,metadata_cols])
    if (display_combined){
      metatext=paste0("Train/Test: ",
                      metadata[1], "/", metadata[2],
                      "\nTrain/Test Target Ratio: ", metadata[3], "/", metadata[4],
                      "\nResampling: ", metadata[5],
                      "\nFeatures: ", metadata[6],
                      "\nGenes Trained: ", combined_training_genes,
                      "\nTest Gene: ", gene
      )
    } else {
      metatext=paste0("Train/Test: ",
                      metadata[1], "/", metadata[2],
                      "\nTrain/Test Target Ratio: ", metadata[3], "/", metadata[4],
                      "\nResampling: ", metadata[5],
                      "\nFeatures: ", metadata[6],
                      "\nTest Gene: ", gene
      )
    }
    
    #print(metatext)
    
    grob <- grobTree(textGrob(metatext,
                              x=0.01,
                              y=0.90,
                              hjust=0,
                              gp=gpar(col="black")
    )
    )
  }
  ggplot(data=plot_data, aes_string(x="Model_name",
                                    y=selection, 
                                    fill="source_data",
                                    group=selection) ) +
    geom_bar(stat="identity"
             , width = 0.75
             , position=position_dodge2(padding=0.1, preserve='total', reverse=TRUE)
    ) +
    coord_cartesian(ylim = c(y_coord_min, 1)) +
    scale_fill_manual(values = c("BT" = "#605ca8",
                                 "CV" = "#bebddb") ) +
    #guides=guide_legend(reverse=TRUE) +
    annotation_custom(grob) +
    # doesn't work with plotly but looks nice :-( 
    geom_label(aes_string(label=selection),
               position=position_dodge(width = -0.75),
               #position=position_dodge2(padding=0.1),
               vjust = 1.5,
               alpha=0.75,
               fill="white"
               
    ) +
    
    # works with plotly but i can't figure out the background yet
    # geom_text(aes_string(label=selection, group=selection),
    #           position=position_dodge(width = -0.75),
    #           vjust = 1.5,
    #           alpha=0.75,
    #           
    # ) +
    
    # add little numbers for the BT bars only
    labs(x="",y=paste(selection,"Score")) +
    theme(
      axis.text.x = element_text(angle = 90),
    )
  # ggplotly()
}