8erberg/spatially-embedded-RNN

Dataset

Closed this issue · 1 comments

nshervt commented

Hello,

I have a question regarding the data generator in your repository. I noticed that you defined eight possible scenarios for the combination of goals and choices, i.e.,

self.mazesDic = {'goal': {0: 9, 1: 9, 2: 19, 3: 17, 4: 17, 5: 7, 6: 19, 7: 7},
                            'ChoicesCategory': {0: 'ul', 1: 'rd', 2: 'ld', 3: 'rd', 4: 'ul', 5: 'ur', 6: 'lr', 7: 'lr'},
                            'NextFPmap': {0: 'u', 1: 'r', 2: 'd', 3: 'd', 4: 'l', 5: 'u', 6: 'r', 7: 'l'}}

The dataset is then randomly sampled from these scenarios for any number of times (number_of_problems).

However, your eight scenarios only cover some possible combinations. Specifically, the up/down ('ud') choice combination is not presented in any trial. Could you please clarify why this combination is not included in the scenarios? Also, you have referred to the file MazeMetadata.py, which is not in this repo. Where can I find this file?

8erberg commented

Hi @nshervt -- thank you for your question!

That is correct. We explain more on the details of the task in the paper, see 'Methods: Task paradigm' (https://www.nature.com/articles/s42256-023-00748-9#Sec9).

In essence, for one of the control analysis described in the supplementary materials, we vary the task difficulty. The harder task used for additional controls uses all combinations and the baseline is a reduced set. There was no particular logic behind choosing this specific subset of trials.

You are correct, MazeMetadata.py is not included here -- this was actually just a helper line for me during development and arguably could have been deleted here to avoid confusion. As the paper above outlined, the task here is essentially a simplified version of a multistep task which we use in an experiment with non-human primates. MazeMetadata.py is the file which describes the full multistep experiment in primates and the line in the docstring is a note of how I extracted the specific set of trials from the full monkey experiment. We hopefully will have a preprint with the full monkey experiment online within the first half of 2024 but if it is relevant for your ongoing work, I am happy to share the full outline of the task with you already!

I hope this answers your question -- happy to add details to aspects which remain unclear!

  • 👍1