draft efficient mask tiling

Author: ryanccarelli

pathml/core/slide_data.py

@@ -133,6 +133,18 @@ def generate_tiles(self, shape=3000, stride=None, pad=False, level=0):
             n_chunk_i = (i - shape[0]) // stride_i + 1
             n_chunk_j = (j - shape[1]) // stride_j + 1
 
+        '''
+        # TODO: check if this keeps order?
+        coordinates = np.array([(x*stride_i,y*stride_j) for x in range(n_chunk_i) for y in range(n_chunk_j)])
+        for mask in self.masks.h5managers.h5['masks'].keys():
+            m = self.masks[mask]
+            first = np.array_split(m, axis=0) 
+            masktiles = []
+            for outs in first:
+                second.append(np.array_split(outs, axis=1))
+            masks = {mask : second}
+        '''
+
         for ix_i in range(n_chunk_i):
             for ix_j in range(n_chunk_j):
                 coords = (int(ix_j * stride_j), int(ix_i * stride_i))

pathml/datasets/peso.py

@@ -16,6 +16,9 @@
 from pathml.core.slide_classes import HESlide
 from pathml.core.masks import Masks
 
+import cProfile
+import pstats
+
 class PesoDataModule(BaseDataModule):
     def __init__(self,
             data_dir,
@@ -67,6 +70,10 @@ def _download_peso(self, download_dir):
         for trainingwsifolder in trainingwsifolders:
             for file in os.listdir(Path(download_dir)/Path(trainingwsifolder)):
                 if file.endswith('.tif'):
+                    
+                    profile = cProfile.Profile()
+                    profile.enable()
+
                     name = '_'.join(file.split('_')[:-1])
                     maskpath = Path(name+'_HE_training_mask.tif') 
                     mask = HESlide(filepath = str(Path(download_dir)/Path('peso_training_masks')/maskpath), name = name)
@@ -85,7 +92,10 @@ def _download_peso(self, download_dir):
                             threshold = 30, outer_contours_only = True)
                         ])
                     # TODO: choose tile size
-                    wsi.run(pipeline, tile_size=3000)
+                    wsi.run(pipeline, tile_size=250)
+                    profile.disable()
+                    ps = pstats.Stats(profile)
+                    ps.print_stats()
                     wsi.write(str(Path(download_dir)/Path('h5')/Path(name+'.h5')))
                     os.remove(str(Path(download_dir)/Path(traininwsifolder)/Path(file)))
                     os.remove(str(Path(download_dir)/Path('peso_training_masks')/maskpath))