parse_and_allocate.R

## Parses a roster page saved from Canvas and allocates students to groups within discussion sections

## To use:  
## 1. Navigate to the People page in a Canvas course using your browser
## 2. Right-click on the page and select "Save as"
## 3a. `Rscript parse.R path/to/roster.html`
## -OR-
## 3b. Save it to `data_folder` as `roster.html`. `Rscript parse.R` or run in RStudio. 

## Setup ----
library(tidyverse)
library(rvest)
library(assertthat)
library(randomizr)

group_size = 5 ## Target size for groups

data_folder = file.path('..', 'data')
out_folder = file.path('..', 'out')
if (!dir.exists(out_folder)) {
    dir.create(out_folder)
}

## If R is not running interactively, get path to roster file from command line
args = commandArgs(trailingOnly = TRUE)
# message(args)
if (!interactive() && length(args) > 0) {
    roster = args[[1]]
} else {
    roster = file.path(data_folder, 'roster.html')
}
# message(roster)

## Load HTML ----
raw = read_html(roster)

## Parsing ----
dataf = raw %>% 
    ## Extract roster table
    html_table() %>% 
    .[[1]] %>% 
    ## Convert to a tibble, w/ nicer names
    as_tibble(.name_repair = 'universal') %>% 
    rename_all(tolower) %>% 
    ## Columns we care about
    select(name, login.id, section, role) %>% 
    ## Need to parse `section` and `role`: 
    ## Can have multiple entries, corresponding to lecture and different discussion sections
    rowwise() %>% 
    mutate_at(vars(section, role), 
              ~ str_split(., '[\n ]+')) %>% 
    ## Unnest; filter down to only students
    unnest(c(section, role)) %>% 
    filter(role == 'Student') %>% 
    select(-role) %>% 
    ## Split section into `lecture` and `discussion` columns
    separate(section, into = c('type', 'number')) %>% 
    pivot_wider(names_from = type, 
                values_from = number) %>% 
    rename_all(tolower)

## Data validation ----
## All lecture values are non-empty
assert_that(all(!is.na(dataf$lecture)), 
            msg = 'missing value for lecture')
## All discussion values are non-empty
# assert_that(all(!is.na(dataf$discussion)),
#             msg = 'missing value for discussion')
# dataf %>% 
#     filter(is.na(discussion))

## No duplicated rows
assert_that(all(!duplicated(dataf)), 
            msg = 'duplicated rows')
## No duplicated login IDs
assert_that(length(unique(dataf$login.id)) == nrow(dataf), 
            msg = 'duplicated login IDs')

## Group allocation ----
set.seed(2020-08-24)
## Students in each discussion section
num_sections = dataf %>% 
    count(discussion) %>% 
    filter(!is.na(discussion)) %>% 
    nrow()
num_arms = ceiling(nrow(dataf) / group_size / num_sections)
allocation = dataf %>% 
    filter(!is.na(discussion)) %>% 
    mutate(group = block_ra(discussion, num_arms = num_arms), 
           group = str_c(discussion, '-', group))

## Validation: All groups within 1 of the target size
allocation %>% 
    count(group) %>% 
    pull(n) %>% 
    {. - group_size} %>% 
    abs() %>% 
    {. <= 1} %>% 
    all() %>% 
    assert_that(msg = 'group too far from group_size')

## Output ----
write_rds(allocation, file.path(out_folder, 'roster.Rds'))
write_csv(allocation, file.path(out_folder, 'roster.csv'))