search

MLD3 / DCEM

[ICML 2024] From Biased Selective Labels to Pseudo-Labels: An Expectation-Maximization Framework for Learning from Biased Decisions
MIT License
2 stars 0 forks source link

Python Version and Crash? #1

Open james-hx-song opened 2 months ago

james-hx-song commented 2 months ago

I am trying to run the code specified in the "legacy" folder on my local macOS, but it seems like some of the packages that you used are for an older version of python. I am currently running on python 3.11; what version did you use?

Also, on line 77 in experiment.py, SEPSIS_PATH = os.path.join(SEPSIS_DICT[socket.gethostname()], SEPSIS_FNAME) gives me an error, and the SEPSIS_DICT constant contains server names; I was wondering if I am supposed to add my own local host name into the dict.

tchang1997 commented 1 month ago

Hi James,

Thanks for your interest in our approach! The legacy folder is for reproducing our results as reported in the paper. The experiments were run on Python 3.9.16; I'll add the minor version to the README.

Good catch on the paths as well — thanks for pointing this out, and I apologize for not catching this earlier —SEPSIS_DICT stores paths for data based on server hostnames, which we used on our machines. But you're correct that this would lead to an error. You can either add your own local host name; alternatively, I've added some commented-in slots where you can set your own path (w/o worrying about the hostname).

In any case, I added some warnings to describe this behavior. Note that if you're interested replicating in the sepsis results, you will need to regenerate the sepsis cohort as well. There are some details in the README for doing so; please reach out with more questions if you'd like.

Are you able to try out the version in this new branch? If this resolves your issues, let me know and I'll merge the changes. Feel free to pull-request on that branch as well if you're interested as well!

Best,

Trenton