Closed Expertium closed 1 month ago
Okay, well, I don't know how to modify multiple files in the same PR https://github.com/open-spaced-repetition/srs-benchmark/pull/114
I just realized that there is a problem. If in algo1.jsonl the data is like this:
user1: 0.5
user2: 0.4
user3: 0.3
user2: 0.2
But in algo2.jsonl the data is like this:
user1: 0.5
user3: 0.3
user3: 0.4
user2: 0.2
The pairwise comparison will be inaccurate. In other words, everything has to be sorted by user
. I'm currently trying to fix it, I'll either make another PR or ask for your help later.
I added a new table to visualize performance of each algorithm. It shows the % of users, for whom algorithm A (row) has lower RMSE than algorithm B (column). For example, 95% means that for mor users, algorithm A results in lower RMSE, and only for a small minority of users, algorithm B results in lower RMSE. I'll also edit the README.