-
Notifications
You must be signed in to change notification settings - Fork 7
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Implementation of reconstruction error ? #37
Comments
Hello Elizabeth, thanks a lot for your tutorials and the good opportunity to reproduce the results of the paper, sorry in advance for the long message. This answer includes :
Don't hesitate if you have any questions, or if you have ideas to improve documentation with what seemed unclear for you. Feedbacks after running your code :
Sidenotes :
With these changes I obtain results that are fairly comparable with the paper. (code below) Full implementation includes:
|
This has been super helpful, thank you !! I've updated my code based on your feedback 😸 Very much appreciated ! I do still have one small trouble I was hoping you could help me track down. I ran my adaption of your code (as well as the exact code you included above), and I'm getting slightly different results for I think these differences are likely coming from one of two places; either (1) my running without subjects 3 and 10 (since these aren't distributed on OSF), or (2) a difference on dependency versions. Here's what I'm currently running the replication with: astroid==2.3.1 Does anything stand out to you ? EDIT: Or, option (3), I'm doing something else wrong entirely ! Any and all feedback is welcome, here. Thanks so much again ! |
This figure comes from the code above ?!😱 |
My running of it 😅 I did add two missing imports so it would run ( |
Glad you found it, I was about looking into this ! It's a bit worrying that these can change so much just based on dependencies. So it's mostly but not only optimal transport, all methods seem to yield different results while all the front-end libraries seemed to be up-to-date. I'm a bit out of ideas here. @bthirion maybe ? |
Ok #40 was merged as consequence of this issue. There is still the dependency question that stays open but i'm not sure we'll debug that quickly. I get back to it when I have time. I close this in the mean time. Thanks a lot for the PR |
Thanks again for making this toolkit !
To understand it a little better, I'm trying to reproduce the Figure 4a from your recent paper introducing fmralign; you can see the implementation I have here, or launch it remotely with Binder.
I can't seem to get my distributions for
ridge
orscaled orthogonal
alignment to match what you have in Figure 4a. I think the most likely explanation is simply that I'm not implementing the formula for reconstruction error appropriately.Do you have an implementation of this already available ?
The text was updated successfully, but these errors were encountered: