Errors in generating cell embeddings

[HelloWorldLTY]

Hi, thanks for your great work. I tested the cell embeddings generation function in a new dataset, and found an error:

File /gpfs/gibbs/pi/zhao/tl688/CellPLM/tutorials/CellPLM/utils/data.py:19, in sparse_scipy_to_tensor(x)
     18 def sparse_scipy_to_tensor(x: scipy.sparse.csr_matrix):
---> 19     return torch.sparse_csr_tensor(x.indptr, x.indices, x.data, (x.shape[0], x.shape[1])).to_sparse().float().coalesce()

RuntimeError: crow_indices.shape[-1] must be equal to the number of rows + 1 (=47682), but got 17142

My dataset has 47681 cells and 17142 genes, and it seems that CellPLM identify my genes as cells. Is it possible to address it? Thanks.

I have checked that this dataset contains non zero gene expression.

[wehos]

Hello, did you convert your dataset from a dense matrix to a csr sparse matrix? For example, adata.X = scipy.sparse.csr_matrix(adata.X). This should be crucial to the model.

I suppose there is an issue with the data format, but I'm surprised that the format checking codes didn't throw an exception.