Open KirstensGitHub opened 9 months ago
Hey @KirstensGitHub the issue is that you're using pre-filtered data rather than re-filtering it. The dataset is too large to store in GitHub, so when you run the analysis notebooks (e.g., analysis.ipynb
), the get_data
function (defined in helpers.py) downloads the dataset to a newly created data
folder.
Once you've run the main analyses (to generate the original figures in the paper), you'll see a file gaze_data.pkl
that gets created in that same data
folder. Because that filtering process can take a while to complete, the next time the data get loaded/filtered, it uses that pickled file rather than re-computing everything. If you don't delete the existing file or change the filename where those pre-computed filtered data are stored (inside load_data
, defined in helpers.py), you'll just end up loading the old results (so nothing will change, even if you change the filtering parameters).
If you only want the raw data, without any filtering, you'd need to modify the load_data
function.
It's of course always possible that I've messed something up, so feel free to dig further and/or ask other questions! I do think the code is organized pretty nicely, so I'd advise against any substantial changes to the organization or how the code is written.
I also see you've updated the image sizes, which are used to compute the gaze intersections with the stimuli. If the new values are correct, you'll definitely want to re-compute the intersections with those updated sizes. That said, I remember having a few conversations about the image sizes, and we had double checked those numbers before running the analyses last time. I think those numbers also matched our prior submissions. Did you uncover new info there?
Hi Jeremy,
I've been looking over the AM code and paper this weekend. I started by reviewing my high-level changes to the code and text. However, the high-level code changes I made were leveraging the latest helper functions. Upon review, I wonder if some of the helper functions are working differently than intended.
For example, I can download the latest repository from CDL github, make no changes, and run the analyses.ipynb notebook. This gives me the figures and results in the paper.
I can then make one change to the notebook: substitute the sustained / variable data with the sustained_unfiltered / variable_unfiltered data. I can then run the notebook again. This gives the same figures and results for all of the memory analyses.
It looks like this happens because the memory data never gets filtered.
The helper functions remove presentation trials where the person looked at the image. They do not remove the corresponding memory trials with those images.
Here are some quick & easy checks:
To make sure we're on the same page before proceeding, I've pushed a minimalist batch of updates. They achieve the following:
Let me know if these make sense or if anything looks off. I think it will be important to get these fundamentals cleared up before I pull request anything else. Thanks and happy 2024!