messcode / stamarker Goto Github PK

Hi,

Thanks for such nice work! I meet a problem while running your Mouse-cerebellum.ipynb tutorial. More specifically, it is about the second box under Visualize spatial domains. The error is as follows.

TypeError: '(slice(None, None, None), (0, 1))' is an invalid key

A screenshot of the error is attached as well. After some googling, it might be about some slicing error, but I could not locate the culprit. I wonder if you know what could be the reason?

Thank you!

Is it necessary to calculate both row and col_linkage since the matrix cmat is symmetric?
col_linkage should be identical to row_linkage, and the function hierarchy.linkage is very slow for larger number of observations.

Also, col_linkage is used only for creation of the image, it is not used in the rest of the code.

Calling zip and itertools to only create pairs of index for symmetrical matrix seems to be too slow. You can use numpy meshgrid function.

row_indices, col_indices = np.meshgrid(indices, indices, copy=True, sparse=False, indexing='ij')
And then just unravel the index matrices:
mat[row_indices.ravel(), col_indices.ravel()] = 1

Please provide some information on writing the model.yaml and trainer.yaml

When using only cpu the tutorial code
if not torch.cuda.is_available():
config["stagate_trainer"]["gpus"] = None
config["classifier_trainer"]["gpus"] = None
creates an error in pipeline.py
if config["stagate_trainer"]["gpus"] > 0:
torch.cuda.empty_cache()
because None cannot be compared to int.
I suggest setting value to 0 instead of None.

I'm struggling to reproduce some results from the paper. I'm running stamarker on the human visium 151507 sample, by adapting code from the tutorial, but I end up with domains that look like this:

Do you have any advice as to what I might be doing wrong? Are there parameters that need to be different than what's shown in the example?

Sklearn and skmisc should have full names (scikit-learn and scikit-misc) so requirements.txt can be automatically run for installation.
Missing torch-sparse package and its version.
Missing scanpy package and its version.
Missing upsetplot package and its version.