steam_data_science

A collection of data science python utilities for analyzing data from the video game platform Steam.

Getting Started

You'll need two things here to start analyzing data: a Steam API key from here and a user's steam64 id, which can be found here. Otherwise, for a given user's steam profile page, if you inspect the page source, you can find your 17 digit steam id key by searching for "g_rgProfileData". An example would look like:

g_rgProfileData = {"url":"https:\/\/steamcommunity.com\/profiles\/**76561197969025704**

Features

build_user_profile(api_key, steam_ids_list)

• Returns a dataframe of a user's (or list of users) games, those games playtime percentiles, and the user's computed "fan rating" engagement score relative to the rest of the population for that game.

compute_fan_rating(data)

• For a given game and its quartiles of playtime data, check the user's playtime against the distribution and compute what percentile the user is compared to the rest of the game's playing population.

data_manager(appid, data_type)

• For a given app and data type like "percentile", "review", or "tag", check against the offline files to see if that data is already stored. If not, retrieve it for faster offline processing.

get_review_data(appid)

• For a given steam app, get its positive and negative review counts for all languages and purchase types.

get_appid_tags(appid)

• For a given steam app, get the tag names, the counts of how many times they've been attributed, and what proportion of the total they make up that app's tags.

get_users_games(api_key, steam_id)

• Get a user's list of games, playtime data, and other info from the IPlayerService and dump it into a pandas dataframe.

get_playtime_percentiles_for_app(appid)

• For a given steam app, gently scrape the howlongis.io site for the app's playtime quartile data.

wishlist_analyzer(wishlist)

• Takes in a wishlist dictionary from the steam page source code and returns a pandas dataframe with review scores, recommended magnitude rank (lower is better), and indie game category levels (lower is more indie).

Example Usage

This example will generate a user profile of engagement data based on individual total playtime of apps in minutes. It can be extended to multiple users by adding additional steam ids to the list:

api_key = '443E...A4B'
steam_ids_list = ['76561197969025704']
user1 = build_user_profile(api_key, steam_ids_list)

If no data for an appid is found, the data is scraped and stored locally to avoid re-scraping for future analyses. Each scrape happens simultaneously so if an appid doesn't have tag, percentile, or review data, all of them are acquired simultaneously but on a 2 second cooldown. So if you have 200 apps that aren't in the file already, it will take about 7 minutes to fully process, but each subsequent re-run will be much faster.

This example will generate an attirubtion model fingerprint based on the user profile you've developed:

user1_fingerprint = compute_profile_fingerprint(user1)

This example line will score an individual game for predicted engagement based on the figerprint data you've developed:

game_tag_scorer('427520', user1_fingerprint[['name','attr_prop_sum']])

This example block will score a list of games for potential engagement based on the fingerprint data you've developed:

candidates_list = ['505460','632360','361420','548430']
candidates_data = []
for i in candidates_list:
    candidates_data.append(game_tag_scorer(i, user1_fingerprint[['name','attr_prop_sum']]))
sorted(candidates_data, key=lambda tup: tup[1],reverse=True)