Unable to patch with use_mask=False #101
Description
Creating HEST objects from some of my objects, I have found that masking sometimes misses tissue areas (both using deep and otsu), so downstream patches are not created for those areas based on the defaults of dump_patches:
I was trying to get around this by setting use_mask=False, but it throws an error about valid barcodes:
UnboundLocalError Traceback (most recent call last)
Cell In[32], line 2
1 new_patch_size = 224 # size of patches used as input for UNI and Virchow2
----> 2 st.dump_patches(patches_dir,
3 target_patch_size=new_patch_size, target_pixel_size=0.5,
4 use_mask=False)
File ~/HEST/src/hest/HESTData.py:346, in HESTData.dump_patches(self, patch_save_dir, name, target_patch_size, target_pixel_size, verbose, dump_visualization, use_mask, threshold, coords_only)
343 if mask is not None:
344 valid_barcodes = barcodes[patcher.valid_mask]
--> 346 patcher.to_h5(h5_path, extra_assets={'barcode': valid_barcodes})
348 if dump_visualization:
349 patcher.save_visualization(os.path.join(patch_save_dir, name + '_patch_vis.png'), dpi=400)
UnboundLocalError: local variable 'valid_barcodes' referenced before assignment"
I also tried just not masking for the sample, and not specifying use_mask, but this led to:
Exception: No tissue segmentation attached to this sample, segment tissue first by calling segment_tissue() for this object
Briefly, I have run my samples with autoalignment upstream- though I have /spatial/ output from cellranger, I do not have alignment.json available, and providing the spatial_coord_path was resulting in misalignment of spots, so I switched to auto.
Is it possible to ignore the masking, and force patching around every spot in st.adata.obs.index?