Open HelloWorldLTY opened 10 months ago
There is a following-up question for this step:
It seems that there still exists errors and warnings for this step. Moreover, for running a spatial datasetsw ith 4,000 cells and 150,000 genes, cellphonedb is obviously faster than spatialDM (10 mins vs 16 hours), which is not consistent with the efficiency plot shown in the readme file. Are there any approaches to acclecrate spatialDM? Thanks.
sdm.weight_matrix(adata, l=120, cutoff=0.2, single_cell=False) # weight_matrix by rbf kernel The parameter ‘l =120’ here might solve this problem.
Also, I found that the parallel parameter ‘nproc=8’ can significantly reduce the processing time.
Hi, I met same problem here even if I tried both l=120 or l=1200. I think there is lack of information for us to choose a good initial value of l and cutoff. Moreover, I need to set single_cell=True because I am handling single-cell data (like MERFISH).
Also I tried to set nproc = 8, the time usage is the same as nproc =1.
@HelloWorldLTY The
flat not found error
is caused by applying numpy methods on csr_matrix. The following code should solve it:
if singlecell:
# Convert csr_matrix to lil_matrix for efficient row operations
rbf_d_lil = rbf_d.tolil()
# Set diagonal elements to zero
rbf_d_lil.setdiag(0)
# Convert back to csr_matrix if needed
rbf_d = rbf_d_lil.tocsr()
Hi, I met same problem here even if I tried both l=120 or l=1200. I think there is lack of information for us to choose a good initial value of l and cutoff. Moreover, I need to set single_cell=True because I am handling single-cell data (like MERFISH).
Also I tried to set nproc = 8, the time usage is the same as nproc =1.
I've been trying with VisiumHD data which now can have from 150k to 650k + spots (barcodes) so I think at this stage SpatialDM isn't quite compatible with single-cell level resolution for spatial data unfortunately
Hi, I met a bug when inferring cci in image-based spatial data:
Could you please take a look? Thanks.