<-- home

Fantasy Hockey with rvest and purrr

rvest and purrr are wonderful bedfellows. The packages share the underlying tidyverse API. And it feels simple and almost natural to combine them when scraping the web.

Here is a slimmed down and worked recipe of how to leverage rvest and purrr in Fantasy Hockey.


Step 0. Load packages.

library(tidyverse)
library(rvest)
library(purrr)
library(stringr)


Step 1. Find a data source.

I’m going to use Fantasy Sports Portal for this example.

Step 2. Figure out the CSS selectors for the data.

SelectorGadget makes this dead simple, like so:

Step 3. Fetch the data elements.

I like to put everything in tibble as soon as possible and use stringr to adjust the url for the different position pages. You’ll notice that I’m only grabbing name and goals. Feel free to grab whatever!

p_fetch <- function(position = "C") {
    
    url <- str_c(sep = "", 
        "https://www.fantasysp.com/projections/hockey/weekly/",
        position)
    
    page <- read_html(url)
    
    names <- page %>%
        html_nodes("td:nth-child(2)") %>% 
        html_text()
    
    goals <- page %>% 
        html_nodes("td:nth-child(4)") %>% 
        html_text()
    
    df <- tibble(name = names, goals)
    
    return(df)
}


Step 4. Iterate through each page.

Instead of writing a for loop, I like to use pmap from purrr to iterate through the Centre, Left-Wing, Right-Wing and Defense position projection pages (I left out the Goalies for obvious reasons).

p_pull <- function() {
    
    params <- tibble(position = c("C", "LW", "RW", "D"))
    
    df <- params %>% 
        pmap(p_fetch) %>% 
        bind_rows()

    return(df)
}


Step 5. Clean and format the projection data.

This is a pretty janky use separate but it works to get everything into a format that I like.

p_clean <- function() {
    
    df <- p_pull() %>% 
        separate(name, 
            into = c("junk", "first", "last", "meta"), 
            sep = "(?=[A-Z][a-z])|(?<=[a-z])(?=[A-Z])",
            fill = "right", 
            extra = "merge") %>% 
        separate(meta, into = c("team", "position"), sep = "\\s") %>% 
        mutate(name = str_c(first, last, sep = "")) %>% 
        mutate(goals = as.numeric(goals)) %>% 
        drop_na() %>% 
        mutate(length = str_length(team)) %>% 
        filter(length <= 3) %>% 
        select(name, team, position, goals)
    
    return(df)
}

df <- p_clean()

At this point you should have a nice clean tibble/dataframe with every player, their position, team, and their projected goals for this week. I could stop here, but I wanted to go a little further with a value over replacement player (VORP) calculation.

Step 6. Calculate a replacement player for each scoring position.

I’m using pmap again to pump through each position to get the mean value for the top X players. It’s a little overkill, but really flexible.

p_replacement <- function(pos, slots) {
    
    rp <- df %>% 
        filter(position == pos) %>% 
        arrange(desc(goals)) %>% 
        filter(row_number() <= slots) %>% 
        group_by(position) %>% 
        summarise(goals = mean(goals))
    
    return(rp)
}

p_vorp <- function() {
    
    # slots depend on how many position players start for each team
    # if there are 10 teams and 2 LW per team then slots -> 10 * 2 = 20
    
    params <- tribble(
        ~pos, ~slots,
        "C", 20,
        "LW", 20, 
        "RW", 20, 
        "D", 20)
    
    rp <- params %>% 
        pmap(p_replacement) %>% 
        bind_rows()
    
    return(rp)
}


Step 7. Calculate the VORP for each player.

Simple join at this point…

replacement <- p_vorp()

# calculate value over replacement player

vorp <- df %>% 
    left_join(replacement, by = "position") %>% 
    mutate(goals_vorp = goals.x - goals.y) %>% 
    rename(goals = goals.x, goals_rp = goals.y) %>% 
    select(-goals_rp) %>% 
    arrange(desc(goals_vorp))


And that’s it!