Suggestion for pro-monitoring program

[John Fairbairn]

I wonder whether someone would be willing implement a (?simple) program to monitor how closely pros follow the bots.

What I envisage is that a pro game can be fed to a bot and it goes through every move, keeping track of how often the pro makes a move within X% (adjustable by the user) of the bot's best move.

The output will simply be something like a tabulation for each player of how many moves up to a certain point (again adjustable by the user) were within that percentage.

There was supposed to be something like this available in the Elf data dump but I think it got lost in the morass of data. Maybe someone else has done it and I missed it? It can be done by hand, of course, but that's so tedious. I think having a running "within X%" score while watching games would be a big plus for spectators.

One possible variation would be a measure of how far away from the bots a player is.

[macelee]

Yikeweiqi has this in its Hawk Eye service for ages. Writing a program to calculate such statistics is the easy part. Generating the data is the hard part. With the existing ELF data, the statistics can be extracted easily if you know how to parse SGF files with tree structures.

[macelee]

These images show how such statistics is displayed in an app. Well, playing more AI moves doesn't necessarily translate into winning games, as shown by this example - black played more AI moves (66.2% times vs white's 60.9%) but actually lost the game.

IMG_3441.PNG (693.54 KiB) Viewed 3519 times

IMG_3442.PNG (649.64 KiB) Viewed 3519 times