poldrack
commented
2 years ago tested this out. I found that while the in-sample models could fit well, at least for the one example fly the out-of-sample prediction accuracy was basically zero. happy to revisit if we can find a better thing to predict or a model that we expect to work.
The accuracy of movement prediction from neural signals could be a useful quality control measure. a first pass was implemented in notebooks/movement_decoding.ipynb - this should be converted to a script and included in workflow.