ValueError: cannot convert float NaN to integer
nbounoua opened this issue · 9 comments
Hello,
I am running into issues when running ICA-Aroma on a subset of the same. Specifically, an error occurs in Step 2 prior to the data denoising step regarding plots and distributions. The error message reads: ValueError: cannot convert float NaN to integer (see screenshot below). We updated to the most recent icaaroma version says that it’s centos7 and that the requirements were setup fine.
Also, I tried to re-run with the --no plot option and it seems to have moved on to the data de-noising step and creates the file but gives me these errors
Hello,
I just wanted to repost these issues when running ICA-Aroma on a subset of the same. Specifically, an error occurs in Step 2 prior to the data denoising step regarding plots and distributions. The error message reads: ValueError: cannot convert float NaN to integer (see screenshot below). We updated to the most recent icaaroma version says that it’s centos7 and that the requirements were setup fine.
First idea is that there might be NaN's in your original input image.
We checked and didn't find any NaNs the original input file or the major outputs in the Melodic directory. There's NaNs in the feature scores and classification overview files, but that's the only place we're seeing them.
Do I understand it correctly that it works fine for some subjects, but not for others?
Another update: We updated the following site-packages requirements for python/ICA_AROMA and we no longer receive the Float NAN to integer error: future, matplotlib, numpy, pandas, & seaborn. However, we are still getting some errors, despite completion of the data denoising step. The denoised file looked good, but we weren't sure if we should proceed.
Is it okay that we updated these packages? If so, w
ould we need to run all participants with these updates?
yes, these packages should not impact the classification. To be on the safe side: yes, I would rerun all participants. Yet given your error messages and the fact that you get them for some, but not all, participants, still makes me wonder if those participants have some 0 data where that would not be expected...
Thanks for the feedback. I share the same concern, do you have suggestions on where I should look for 0s?