Awesome work! Here's 57 million run history records.

[JakeRabinowitz]

Hi there, this is super impressive work! I run the Slay the Spire internal metrics dashboard, and was running a run history export script from October 2018 -> July 2019. Over that time I accrued about 57 million runs worth of metrics, with the idea that they could be useful for projects like this one. I turned off the data export a year ago, but I'll turn it on again for a while to get some more up to date data for you.

Here's a link to the run files I have now: https://drive.google.com/drive/folders/1c7MwTdLxnPgvmPbBEfNWa45YAUU53H0l?usp=sharing

There are over 35,000 files, each contains ~1,600 runs in a json list. The file names are [timestamp]#[runcount].json.gz. I recommend you use a faster json library than the default 😉 I find ujson to be pretty effective.

[alexdriedger]

Thank you so much! I'll try training the model with the new data

[JakeRabinowitz]

Oh just fyi: If you're trying to split runs within a file into training/test sets, you need to select runs randomly instead of picking a range of runs for the test set, since the runs files are organized/grouped by the DB. If you just designate entire run files as test data, that should be fine since they're just exports of the past 10 minutes worth of runs.