TheorySpace

Experiments

Experiments

Small computational studies, one methods question each. I write the protocol down before I look at any results, and I report what didn't work.

All experiments

Nothing here yet.

How each one is set up

Question
One question, narrow enough that a result could settle it.
Hypothesis
What I expect to happen and why, written before I run anything.
Data
A public dataset, or a simulation with the generating process spelled out.
Methods and baselines
The method I'm testing, next to the simplest thing that could plausibly work.
Metrics
Picked in advance. Usually calibration, predictive error, parameter recovery or cost.
Findings and limits
Results, figures and code, plus the conditions I think they depend on.

Suggest one

If there's a methods question you'd like to see tested, or you think one of my results is wrong, please tell me. Being told I'm wrong is usually how these get better.