Merge branch 'main' into upsample

dacapo/experiments/datasplits/datasets/arrays/binarize_array_config.py

@@ -55,12 +55,13 @@ def array(self, mode="r") -> Array:
         assert num_channels is None, "Input labels cannot have a channel dimension"
 
         def group_array(data):
-            out = da.zeros((len(self.groupings), *array.physical_shape), dtype=np.uint8)
-            for i, (_, group_ids) in enumerate(self.groupings):
-                if len(group_ids) == 0:
-                    out[i] = data != self.background
-                else:
-                    out[i] = da.isin(data, group_ids)
+            groups = [
+                da.isin(data, group_ids)
+                if len(group_ids) > 0
+                else data != self.background
+                for _, group_ids in self.groupings
+            ]
+            out = da.stack(groups, axis=0)
             return out
 
         data = group_array(array.data)

dacapo/experiments/datasplits/datasets/arrays/concat_array_config.py

@@ -4,7 +4,6 @@
 
 from typing import List, Dict, Optional
 from funlib.persistence import Array
-import numpy as np
 import dask.array as da
 
 
@@ -45,18 +44,15 @@ class ConcatArrayConfig(ArrayConfig):
     def array(self, mode: str = "r") -> Array:
         arrays = [config.array(mode) for _, config in self.source_array_configs.items()]
 
+        out_data = da.stack([array.data for array in arrays], axis=0)
         out_array = Array(
-            da.zeros(len(arrays), *arrays[0].physical_shape, dtype=arrays[0].dtype),
+            out_data,
             offset=arrays[0].offset,
             voxel_size=arrays[0].voxel_size,
             axis_names=["c^"] + arrays[0].axis_names,
             units=arrays[0].units,
         )
 
-        def set_channels(data):
-            for i, array in enumerate(arrays):
-                data[i] = array.data[:]
-            return data
-
-        out_array.lazy_op(set_channels)
+        # callable lazy op so funlib.persistence doesn't try to recoginize this data as writable
+        out_array.lazy_op(lambda data: data)
         return out_array

dacapo/experiments/datasplits/datasets/arrays/logical_or_array_config.py

@@ -30,14 +30,14 @@ def array(self, mode: str = "r") -> Array:
         assert num_channels_from_array(array) is not None
 
         out_array = Array(
-            da.zeros(*array.physical_shape, dtype=array.dtype),
+            da.zeros(array.physical_shape, dtype=array.dtype),
             offset=array.offset,
             voxel_size=array.voxel_size,
             axis_names=array.axis_names[1:],
             units=array.units,
         )
 
-        out_array.data = da.maximum(array.data, axis=0)
+        out_array.data = da.max(array.data, axis=0)
 
         # mark data as non-writable
         out_array.lazy_op(lambda data: data)

dacapo/experiments/tasks/predictors/hot_distance_predictor.py

@@ -1,7 +1,6 @@
 from dacapo.experiments.arraytypes.probabilities import ProbabilityArray
 from .predictor import Predictor
 from dacapo.experiments import Model
-from dacapo.experiments.arraytypes import DistanceArray
 from dacapo.tmp import np_to_funlib_array
 from dacapo.utils.balance_weights import balance_weights
 
@@ -394,6 +393,7 @@ def __find_boundaries(self, labels):
         # bound.: 00000001000100000001000      2n - 1
 
         logger.debug(f"computing boundaries for {labels.shape}")
+        labels = labels.astype(np.uint8)
 
         dims = len(labels.shape)
         in_shape = labels.shape

pyproject.toml

@@ -202,6 +202,7 @@ module = [
     "napari.*",
     "empanada.*",
     "IPython.*",
+    "xarray_multiscale.*"
 ]
 ignore_missing_imports = true
 

tests/components/test_gp_arraysource.py

@@ -30,6 +30,5 @@ def test_gp_dacapo_array_source(array_config):
         batch = source_node.request_batch(request)
         data = batch[key].data
         if data.dtype == bool:
-            raise ValueError("Data should not be bools")
             data = data.astype(np.uint8)
         assert (data - array[array.roi]).sum() == 0