Hi, in the original paper by Berman, he states the use of the KL Divergence as the metric to compare two points (wavelet coefficients). Did you achieve any success with using the "symmetric_kl" metric in UMAP? Secondly, I get very poor performance for my behaviour data using the wavelets, even after subsampling the original data. Do you have any suggestions how to improve the same? I have keypoint data (identified by DeepLabCut) at a sampling rate of 30Hz.

Hi @kanishkbjain !

I tried installing motionmapperpy in my own machine, with Ubuntu and python 3.7.13, the python version we used in the collab tutorial, as reported by sys.version)

In order to install motionmapperpy in this python version, I had to pin the version of numpy scipy and scikit-learn as follows in the setup.py

install_requires=[
    "numpy==1.21.6",
    "scipy==1.4.1",
    "matplotlib",
    "h5py",
    "hdf5storage",
    "scikit-learn==1.0.2",
    "scikit-image",
    "easydict",
    "umap-learn"
],

leaving it here in case anyone encounters versioning issues :-)

Hi @kanishkbjain, could you give more details on what the dataset should look like?
Also, details on using poses of multiple mouse videos should be very helpful.

Thanks!

HI! Thank you for sharing this MotionMapper python code. When I ran the demo code (demo.py), I received the following error:

~/motionmapperpy$ python3 demo/demo.py
...
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "h5py/h5a.pyx", line 50, in h5py.h5a.create
OSError: No write intent on file (no write intent on file)

Can anyone please help in sorting out this issue?