LoZscraper

This project is analysis of Legend of Zelda (1986) speedruns from Arcus87. Here I include my Arcus Twitch Vod Scraper (Python), the scaped data, and my analysis (R Markdown)

Background

I got into watching speedrunner's on Twitch and while back. In particular I was got into who was into speedrunning Legend of Zelda (1986), for the NES, a game a played a lot of as a kid.

Watching him, I thought it would be cool to collect some data on how his speed run was progressing, and I had a few questions:

• Which rooms give him the most trouble?
• How big a deal is it to have the "beam-sword"?
• How much better is he getting over time?

I have started to blog about this on Medium which has more descriptive analysis.

Using the scraper

The scraper scrapeLOZ.py was really my first attempt doing a Python project and it's not a pretty thing, but it works. I'm going to document how to use the scraper more fully later but it's not a big priority.

The scraper works on an .mp4 file downloaded from twitch, I used Arne Vogel's Concat but you're free to use whatever tool you like.

• "-v" or "--video" -v='video2scrape.mp4 required this specifies the file to scrape
• "--room_list" --room_list='roomlist.csv' this specifies the room list to use while scraping, if none is specified this will default to '../data/unique_room_list_double_hundo_with_index.csv'
• "--verbose" gives more output
• "-t" or "--start" -t=1000 time in the video to start scraping given in milliseconds
• "-run" -run=1345 this species the run number for the first run that begins after your start time
• "-d" of "--delta" time in milliseconds for advance when initially checking for next room, defaults to 3000

The data

Output will be an sqlite database .db file of the same name as your video.

Todo

I'm happy with the current version for my analysis but I may continue to update planned features in the future.

Planned Features

• Check to see if video file exists
• Check if Run ends on Room 1
• Specify an end time to stop scraping
• Manual mode for scraping, will not need a room list.
• No save / continuation feature
• Adjustable time precision, currently a precision of 0.1 s is used for changes
• Migrate from tesseract to opencv template matching for recognition