Steam Portal

This page lists my Github repositories which objective is to analyze Steam data.

I. Store descriptions

1. Similarity

• steam-descriptions: retrieve games with similar store descriptions,
• woctezuma.github.io: visualize, as a graph, games with similar store descriptions.

2. Generative Language Model

• sample-steam-descriptions: sample store descriptions, with the GPT-2 language model.

II. Store banners

1. Similarity

• download-steam-banners: retrieve games with similar banners, using MobileNet (alpha 0.25, input 128x128x3).
• match-steam-banners: retrieve games with similar banners, using MobileNet v3 (alpha 1.00, input 256x256x3).
• steam-CLIP: retrieve games with similar store banners, using OpenAI's CLIP (input 224x224x3).
• steam-DINO: retrieve games with similar store banners, using Facebook's DINO (input 224x224x3),
• steam-DINOv2: retrieve games with similar store banners, using Meta AI's DINOv2,
• steam-I-JEPA: retrieve games with similar store banners, using Meta AI's I-JEPA,
• steam-BiT: retrieve games with similar store banners, using Google's Big Transfer (BiT) (input 224x224x3),
• steam-BEiT: retrieve games with similar store banners, using Microsoft's Bidirectional Encoder representation from Image Transformers (BEiT) (input 224x224x3),
• steam-palette-extractor: retrieve games with similar palettes,

2. Generative Adversarial Networks

• google-colab: train a DCGAN to generate 28x28 store banners,
• steam-stylegan: train a StyleGAN to generate 128x128 store banners,
• steam-stylegan2: train a StyleGAN2 to generate 256x256 store banners.
• steam-stylegan2-ada: train a StyleGAN2-ADA to generate 256x256 store banners.
• steam-lightweight-gan: train a "lightweight GAN" to generate 256x256 store banners.
• steam-stylegan2-ada-pytorch: train a StyleGAN2-ADA (PyTorch) to generate 256x256 store banners.

3. Face detection

• steam-face-detection: detect (and count) faces on Steam store/library banners, for filtering purposes.

4. Search engine

• steam-image-search: search for images on Steam using natural language queries.
• steam-template-matching: perform template matching with Steam store images.

III. Store screenshots

1. Similarity

• download-steam-screenshots: retrieve games with similar store screenshots, using MobileNet (alpha 0.25, input 128x128x3).

IV. Store reviews

1. Generative Language Model

• sample-steam-reviews: sample reviews, with different methods (including GPT-2),
• sample-steam-reviews-with-gpt-2: sample reviews, with the GPT-2 language model. Shorter & simpler code.

2. Stats

• steam-reviews: stats and sentiment analysis of reviews,
• steam-cut: fetch the number of reviews for Steam and non-Steam purchases,
• players-vs-reviews: discover games with an unusual ratio of number of players vs. number of reviews.
• steam-reviews-to-sales: study the relationship between review numbers and sales of games,

V. Scores

1. Rankings

• hidden-gems: rank games to discover "hidden gems". NB: This lead to my contribution to Steam250,
• Steam-Bayesian-Average: rank games, developers and publishers,
• Steam-Labs-Recommender: tweak the Interactive Recommender of Steam Labs,
• hidden-demos: rank games which offer a demo.

2. Community

• metacouncil-goty: ranked-choice voting for Games of the Year Awards on MetaCouncil/Steam in 2018-2020,
• steam-era-goty: ranked-choice voting for Games of the Year Awards on ResetERA/Steam in 2017,
• steam-groups: descriptive stats regarding the libraries of members of a Steam group.

VI. Store tags

• match-steam-tags@PyPI: download and match Steam tags with appIDs,
• steam-tag-mapping: visualize an embedding of Steam tags.
• epic-games-tag-mapping: visualize an embedding of tags on the Epic Games Store.

VII. Miscellaneous

• humble-monthly: analyze the content of Humble Monthly Bundles,
• recent-sales: a crude attempt at estimating Steam sales and revenue,
• steam-hype: query a third-party API to find upcoming Steam games with many followers,
• steam-trade-finder: find Steam trades for a given appID,
• steam-market: find arbitrages on the Steam Market.
• steam-points-shop: find games which have been removed from the Steam Points Shop.
• AmongUs-DAU: compute "Daily Active Users" (DAU) of Among Us from "Concurrent Connected Users" (CCU),
• bundle-split: fetch bundle content, then price bundle split based on prices at keyshop resellers.
• steampy-buy-orders: automatically set buy orders on the Steam Market with steampy.
• steam-next-fest: fetch new demos from the Steam Next Fest, then play them as fast as possible.
• add-free-licenses: automatically add free licenses on Steam with ASF IPC.
• steam-player-level-percentiles: query and analyze percentiles of player levels on Steam.

VIII. Data scraping

1. GameDataCrunch API

• gamedatacrunch@PyPI: an API to download data through GameDataCrunch API.

2. SteamSpy API

• steamspypi@PyPI: an API to download data through SteamSpy API.

3. Steam API

• download-steam-reviews@PyPI: download every review for the game of your choice,
• steampi@PyPI: download app details (including release date) via Steam API, and match names with appIDs,
• steam-api: download a list of all appIDs through Steam API, then download their app details, create an aggregate similar to SteamSpy's (but more exhaustive), finally analyze store attributes and map tags.

NB: With steam-api, store descriptions can also be aggregated for later processing.

4. My own API

i) Heroku endpoints

• heroku-flask-api: serve the matching results through an API built with Flask on Heroku,
• heroku-levenshtein-api: match game names with Levenshtein distance via an API built with Flask on Heroku,
• heroku-clip: classify images through an API built with Streamlit on Heroku,

i) API utils

• steam-svelte-autocomplete: auto-complete game names with Svelte,
• steam-popular-appids: create a short list of popular Steam appIDs,
• steam-curator-release-dates: query Steam API to reveal hidden release dates,

IX. Data snapshots

• steam-store-snapshots: app lists,
• steam-api-data: app details,
• download-steam-banners-data: store banners,
• download-steam-screenshots-data: store screenshots,
• steam-filtered-image-data: filtered library images (vertical banners, logos), and the Steam-OneFace dataset,
• steam-reviews-data: reviews,
• steamspy-data: SteamSpy data snapshot,
• hidden-gems-data: language stats for regional rankings of "hidden gems",
• recent-sales-data: aggregate of SteamSpy data between July 17th, 2017 and February 6th, 2018.

X. Data leaks

• egs-15DaysofGames: datamining of 15 Days of (free) Games on the Epic Games Store (EGS),
• egs-datamining: datamining of unreleased games on the Epic Games Store (EGS),
• epic-games-ratings: Bayesian average ratings of games on the Epic Games Store,
• epic-games-achievements: achievements on the Epic Games Store,
• epic-games-tracker: track the numbers of players and ratings on the Epic Games Store,
• epic-games-promotion-tracker: track upcoming promotional offers on the Epic Games Store,
• epic-games-gift-tracker: track upcoming gift wrappings on the Epic Games Store,
• epic-games-vault-tracker: track upcoming vaulted mystery games on the Epic Games Store,
• epic-games-player-estimates: correlate the numbers of players and ratings on the Epic Games Store,
• epic-games-js: download JavaScript files (cached webpack) from the Epic Games Store,
• geforce-leak: datamining of Nvidia's GeForce NOW (GFN),
• humble-choice-leak: datamining to retrieve leaks for Humble Choice.