Halite Home

Diamond Replay Dump


I am exploring ML strategies for my Halite bot, and for that, I needed to collect a lot of data. So I downloaded all the replays where a diamond-rank bot wins. Here's the zip file of those 1583 replays (unzipped size 3.6 GB, zipped size 283 MB) involving any of the 17 diamond-rank bots. I organized them by game size (e.g. 20, 25, 30, etc.), too. https://www.dropbox.com/s/gdddajc7z23pt8u/replays.zip?dl=0

And here's the JSON metadata for those replays. https://www.dropbox.com/s/ttoojmdrwollslk/replays.json?dl=0

The metadata is structured as an array of objects, each one describing a replay. Here's the schema:

    'replayName': 'ar<seed>-<timestamp>.hlt',
    'winningPlayer': <player number, one-indexed>,
    'gameSize': <game size>



Hey andrewyang96,

To get the data, did you just pull all the most recent replays off of the Amazon server and filter them based on a diamond player winning or did you selectively pull the data from the server? If so, I'd love to hear how you did it. I'm looking to download replays from several specific players and I'm not sure the best way to go about doing this.


You selectively pull replay files. Go to a player's profile to find his ID.

Use the following link to download initial JSON: https://2016.halite.io/api/web/game?userID='+userId+'&limit=50

Use this link to fetch more: https://2016.halite.io/api/web/game?userID='+userId+'&limit=50&startingID=' + startingID

Next, download each .hlt replay file from https://s3.amazonaws.com/halitereplaybucket/ + replayName


Thanks for the quick reply, this was really helpful. Thank you!


Hey hmate9, I just got around to trying this out. I was able to download some .hlt files, though I have two followup questions.
1) What is the startingID?
2) If there a way to filter based on specific bot version?


StartingID is the first replay ID you want to start with. So first you don't have anything, than you use the last replayID in order to be able to 'scroll' through the replays.

I am not sure how to get older bot versions history.