feat: 💥 Add blockwise prediction, callable from cli.

Added ability to call dacapo.predict from command line. This implements a blockwise-parallelized operation, with specific implementation depending on the ComputeContext. This feature came with the development of the basis for the blockwise processing infrastructure within dacapo.

The function signature of dacapo.predict has changed significantly. apply.py and any other functions/scripts referring to it will no longer function and will need to be converted to the new call structure.

dacapo/apply.py

@@ -192,7 +192,7 @@ def apply_run(
         input_array,
         prediction_array_identifier,
         output_roi=roi,
-        num_cpu_workers=num_cpu_workers,
+        num_workers=num_cpu_workers,
         output_dtype=output_dtype,
         compute_context=compute_context,
         overwrite=overwrite,

dacapo/blockwise/__init__.py

@@ -1 +1,2 @@
 from .blockwise_task import DaCapoBlockwiseTask
+from .scheduler import run_blockwise

dacapo/blockwise/blockwise_task.py

@@ -4,13 +4,14 @@
 from typing import Callable, Optional
 from daisy import Task, Roi
 from dacapo.compute_context import ComputeContext, LocalTorch, Bsub
+import dacapo.compute_context
 
 
 class DaCapoBlockwiseTask(Task):
     def __init__(
         self,
         worker_file: str or Path,
-        compute_context: ComputeContext,
+        compute_context: ComputeContext or str,
         total_roi: Roi,
         read_roi: Roi,
         write_roi: Roi,
@@ -21,6 +22,9 @@ def __init__(
         *args,
         **kwargs
     ):
+        if isinstance(compute_context, str):
+            compute_context = getattr(dacapo.compute_context, compute_context)()
+
         # Make the task_id unique
         timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
         task_id = worker_file + timestamp