Now that we have a functional baseline algorithm and evaluation function for a simple study #3, we want to build up some simple algorithms to traverse through the entire decision process.
Give the algorithm access to the scoring function and just greedily select groupings that would minimize difference score (the lower the score the better performing)
This is probably too slow, and if enough people arrive at once, the combinations possible to test are too high.
Perhaps optimize this later on if we want to revisit this?
Greedy algorithm that weighs samples with lower probability greater
Randomly sample from the pool of participants and build a distribution
Compute pdf (kde) over this random space to determine which points are less likely
Resample by inverse pdf
We want to build out the entire testing machinery on simple algorithms so that we easily apply this pipeline to other algorithms.
Alan-Qiao
commented
11 months ago
Now that we have a functional baseline algorithm and evaluation function for a simple study #3, we want to build up some simple algorithms to traverse through the entire decision process.
We want to build out the entire testing machinery on simple algorithms so that we easily apply this pipeline to other algorithms.