02-rfs-for-app.Rmd

---
title: "Fitting Random Forests for WhoseEgg Shiny App"
author: "Katherine Goode <br>"
date: 'Last Updated: `r format(Sys.time(), "%B %d, %Y")`'
output: rmarkdown::github_document
#output: pdf_document
---

```{r setup, include = FALSE}
knitr::opts_chunk$set(echo = TRUE, message = FALSE)
```

This document contains code that fits the three random forest models that will be used in the app: models with invasive carp species grouped into one class and all other species classified into species, genus, and family. The data used to train the model is that used in Goode et al. (2021) to train the augmented models, and the same seed (808) is used, so the models should agree.

Load packages:

```{r}
library(dplyr)
library(randomForest)
library(purrr)
```

Make a list of the response variables: 

```{r}
vars_resp = c(
  "Family_IC",
  "Genus_IC",
  "Common_Name_IC"
)
```

Make a vector of the predictor variables:

```{r}
vars_pred = c(
  "Month",
  "Julian_Day",
  "Temperature",
  "Conductivity",
  "Larval_Length",
  "Membrane_Ave",
  "Membrane_SD",
  "Membrane_CV",
  "Embryo_to_Membrane_Ratio",
  "Embryo_Ave",
  "Embryo_SD",
  "Embryo_CV",
  "Egg_Stage",
  "Compact_Diffuse",
  "Pigment",
  "Sticky_Debris",
  "Deflated"
)
```

Load the prepared egg data and convert necessary variables to factors:

```{r}
eggdata_for_app <- 
  read.csv("../data/eggdata_for_app.csv") %>%
  mutate_at(
    .vars = c(
      "Egg_Stage",
      "Compact_Diffuse",
      "Pigment",
      "Sticky_Debris",
      "Deflated",
      all_of(vars_resp)
    ),
    .funs = factor
  )
str(eggdata_for_app)
```

Function for fitting a random forest model given a response variable, predictor variables, and a dataset (uses the same seed to fit the random forests as Camacho et al. (2019) and Goode et al. (2021)):

```{r}
fit_rf <- function(resp, preds, data) {
  
  # Fit the random forest
  set.seed(808)
  rf <- randomForest(
    data %>% pull(resp) ~ .,
    data = data %>% select(all_of(preds)),
    importance = T,
    ntree = 1000
  )
  
  # Put model in a named list
  rf_list = list(rf)
  names(rf_list) = resp
  
  # Return the named list
  return(rf_list)
  
}
```

Fit the random forest models: 

```{r}
rfs_for_app <-
  map(
    .x = vars_resp,
    .f = fit_rf,
    preds = vars_pred,
    data = eggdata_for_app
  ) %>%
  flatten()
```

Save the random forests:

```{r}
saveRDS(
  object = rfs_for_app, 
  file = "../data/rfs_for_app.rds"
)
```

# Session Info

```{r}
sessionInfo()
```