search

flatironinstitute / CaImAn

Computational toolbox for large scale Calcium Imaging Analysis, including movie handling, motion correction, source extraction, spike deconvolution and result visualization.
https://caiman.readthedocs.io
GNU General Public License v2.0
637 stars 370 forks source link

Very high RAM usage in demo_pipeline.ipynb #1184

Closed salehtahini closed 10 months ago

salehtahini commented 1 year ago

Setup:

  1. Operating System (Linux, MacOS, Windows): Windows 10 / Windows 11 / Ubuntu 22.04

  2. Hardware type (x86, ARM..) and RAM: x86 (Intel core i9-13900k or Xeon gold 6138), RAM: 64GB

  3. Python Version (e.g. 3.9): 3.8 - 3.11

  4. Caiman version (e.g. 1.9.12): 1.19.15

  5. Which demo exhibits the problem (if applicable): demo_pipeline.ipynb

  6. How you installed Caiman (pure conda, conda + compile, colab, ..):

    conda install -n base -c conda-forge mamba
mamba create -n caiman -c conda-forge caiman
conda activate caiman
python /DIR/TO/caimanmanager.py install

  7. Details: For some reason, the Jupyter version of the pipeline demo didn't want to run on a multipage OME tiff file (16-bit, 28000 Frames, 15GB), it will always allocate all of the RAM and crash during the motion correction. I tried splitting the file to 7 parts, the same would also happen. Reducing n_processes to 12 or 8 can help but then the same problem will occur when trying to run Memory mapping. I tried converting the recording to hdf5 and the same issue still persists. Different python versions didn't differ. Running the python file version demo_pipeline.py in CLI doesn't show the same problems. It would run smoothly on all cores and with only 20-30GB RAM usage, only jumping to almost 100% during Memory mapping.

This issue was observed on 2 different workstations, on Windows and on fresh installs of Ubuntu as well.

Side note: caimanmanager.py install always showed the error "ModuleNotFoundError: No module named 'caiman'". This was regardless of in which directory the command was executed. Using python /DIR/TO/caimanmanager.py install helped circumvent this problem. In 3/4 of my trials, it was after installing anaconda (latest version) for the first time.

EricThomson commented 1 year ago

Second problem first: this shouldn't be a problem with the new release -- see new install instructions on the readme quickstart. We really hope this makes installation easier for people. You should be able to just enter caimanmanager install. Please let us know, this has been a long time coming. :octocat:

The other problem -- I am not sure why it would be different from Jupyter versus the module. I wonder if there are different parameters in the two cases that would yield these differences? We often get people asking about differences between the two demos, but in every case I have looked into it, where they appear to be giving different results, the params have ended up being different in subtle ways. Though honestly I'm not sure what would yield this difference (maybe differences in the elastic motion correction params -- strides/overlaps, or maybe the ssub or tsub params in the patches?). These kinds of "why are my ipynb and py files being different" are very hard to figure out a priori I would print out the parameters and look at them one by one.

EricThomson commented 10 months ago

Closing due to inactivity.