### 4.1.1 Shapefiles, one shapefile for each class.  
### Get the names of the training shapefiles
train.files <- list.files(pattern="*train.*shp$")
### Read in the raster image.  Replace name of image file with the name of the image.
image <- "name of image file"
img <- stack(image)
### Create an object to receive the training data
train.full <- NULL
### This loop extracts the training data for each shapefile and adds it sequentially to the training dataset.
for (x in seq_along(train.files)) {     
  ### Read in the a shapefile
  train.locations <- st_read(train.files[x])
  ### Extract values from the image for the shapefile area.
  train.predictors <- raster::extract(img, train.locations)
  ### Combine the data from multiple polygons in the shapefile.  If the data is in a point shapefile instead of polygons, skip this next line.
  train.predictors <- as.data.frame(do.call(rbind, train.predictors))  
  ### Add a column for the class type and populate it.  
  train.by.class <- as.data.frame(cbind(train.predictors, "types" = substr(train.files[x], 1, nchar(train.files[x])-9))) 
  ### Combine the data with data from previous loops.
  train.full <- rbind(train.full, train.by.class)
  ### Make all your response variables numeric (categorical variables will be addressed later)
  train.full <- train.full %>% mutate_each(funs(as.numeric), starts_with(substr(image, 1, nchar(image)-4)))
}


### 4.1.2 One shapefile with multiple classes. 
### Read in the raster image.  Replace name of image file with the name of the image.
image <- "name of image file"
img <- stack(image)
### Read in the training shapefile.  Replace shapefile with the name of the shapefile.  The shapefile should have an attribute named "types" for class designations (e.g., water).
train.locations <- st_read("shapefile")
### Create an object to receive the training data.
train.full <- NULL  
### This loop extracts the training data for each class and adds it sequentially to the training dataset.
for (i in 1:length(unique(train.locations[["types"]]))){
  ### Sequentially extract the name for each unique class.                          
  train.type <- unique(train.locations[["types"]])[i]
  ### Sequentially create a spatial polygon object for each class.
  train.type.locations <- train.locations [train.locations[["types"]] == train.type,]
  ### ### Extract values from the image for the subset of the shapefile area.
  train.by.class <- raster::extract(img, train.type.locations)
  ### Combine the data from multiple polygons in the shapefile.  If the data is in a point shapefile instead of polygons, skip this next line.
  train.by.class <- as.data.frame(do.call("rbind", train.by.class))
  ### Add a column for the class type and populate it.
  train.by.class <- cbind(train.by.class, "types" = as.character(unique(train.locations[["types"]])[i]))
  ### Combine the data with data from previous loops.
  train.full <- rbind(train.by.class, train.full)
  ### Make all your response variables numeric (categorical variables will be addressed later)
  train.full <- train.full %>% mutate_each(funs(as.numeric), starts_with(substr(image, 1, nchar(image)-4)))
}


### 4.1.3 Csv files, one csv file for each class.
### Get the names of the training csv files.
train.files <- list.files(pattern="*train.*csv")
### Read in the raster image.  Replace image with the name of the image.
img <- stack("image")
### Create an object to receive the training data.
train.full <- NULL
### This loop extracts the training data for each csv file and adds it sequentially to the training dataset.
for (x in seq_along(train.files)) {     
  ### Read in the csv file.
  train.locations <- read.csv(train.files[x])
  ### Set the values in the csv data to be coordinates.
  coordinates(train.locations) <- c("x","y")
  ### Extract values from the image for the point locations.
  train.predictors <- as.data.frame(raster::extract(img, train.locations))
  ### Add a column for the class type and populate it.  
  train.by.class <- cbind(train.predictors, "types" = substr(train.files[x], 1, nchar(train.files[x])-9)) 
  ### Combine the data with data from previous loops.
  train.full <- rbind(train.full, train.by.class)
}


### 4.1.4 Csv file with training data already extracted and assembled. 
### Read in the training data, replacing the filename with the name of the csv file.
train.full <- read.csv("my-data-file.csv")