Merge pull request #7 from TimMonko/#4-fix

Batch training and prediction widgets

setup.cfg

@@ -37,6 +37,10 @@ install_requires =
     magicgui
     qtpy
     aicsimageio
+    napari
+    apoc
+    pyclesperanto_prototype
+    dask
 
 python_requires = >=3.8
 include_package_data = True

src/napari_ndev/__init__.py

@@ -3,8 +3,10 @@
 except ImportError:
     __version__ = "unknown"
 
-from ._widget import batch_annotator
+from ._widget import batch_annotator, batch_predict, batch_training
 
 __all__ = [
     "batch_annotator",
+    "batch_training",
+    "batch_predict",
 ]

src/napari_ndev/_tests/test_widget.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-from napari_ndev import batch_annotator
+from napari_ndev import batch_annotator  # , batch_predict, batch_training
 
 
 # make_napari_viewer is a pytest fixture that returns a napari viewer object
@@ -22,3 +22,41 @@ def test_batch_annotator(make_napari_viewer, capsys):
     # read captured output and check that it's as we expected
     # captured = capsys.readouterr()
     # assert captured.out == "napari has 1 layers\n"
+
+
+# def test_batch_training(make_napari_viewer, capsys):
+#     # make viewer and add an image layer using our fixture
+#     viewer = make_napari_viewer()
+#     test_image = np.random.random((100, 100))
+#     viewer.add_image(test_image)
+#     test_thresh = test_image > 1
+#     viewer.add_labels(test_thresh)
+
+#     # create our widget, passing in the viewer
+#     my_widget = batch_training()
+#     my_widget()
+#     # call our widget method
+#     # my_widget._on_click()
+
+#     # read captured output and check that it's as we expected
+#     # captured = capsys.readouterr()
+#     # assert captured.out == "napari has 1 layers\n"
+
+
+# def test_batch_predict(make_napari_viewer, capsys):
+#     # make viewer and add an image layer using our fixture
+#     viewer = make_napari_viewer()
+#     test_image = np.random.random((100, 100))
+#     viewer.add_image(test_image)
+#     test_thresh = test_image > 1
+#     viewer.add_labels(test_thresh)
+
+#     # create our widget, passing in the viewer
+#     my_widget = batch_predict()
+#     my_widget()
+#     # call our widget method
+#     # my_widget._on_click()
+
+#     # read captured output and check that it's as we expected
+#     # captured = capsys.readouterr()
+#     # assert captured.out == "napari has 1 layers\n"

src/napari_ndev/_widget.py

@@ -1,12 +1,17 @@
 """
 neural development (nDev) widget collection
 """
+import os
 import pathlib
 import string
+from enum import Enum
 from typing import TYPE_CHECKING
 
+import apoc
+import dask.array as da
 import napari
 import numpy as np
+import pyclesperanto_prototype as cle
 from aicsimageio import AICSImage
 from magicgui import magic_factory
 from napari import layers
@@ -83,3 +88,123 @@ def saver(image_type, folder_suffix, save_suffix_str):
     saver(image, "_images", save_suffix)
     saver(labels, "_labels", save_suffix)
     return "Saved Successfully"
+
+
+channel_nums = [0, 1, 2, 3, 4]
+PDFS = Enum("PDFS", apoc.PredefinedFeatureSet._member_names_)
+
+
+@magic_factory(
+    auto_call=False,
+    call_button="Batch Train",
+    result_widget=True,
+    image_directory=dict(widget_type="FileEdit", mode="d"),
+    label_directory=dict(widget_type="FileEdit", mode="d"),
+    predefined_features=dict(widget_type="ComboBox", choices=PDFS),
+    channel_list=dict(widget_type="Select", choices=channel_nums),
+)
+def batch_training(
+    image_directory=pathlib.Path(),
+    label_directory=pathlib.Path(),
+    cl_filename: str = "classifier.cl",
+    predefined_features=PDFS(1),
+    custom_features: str = None,
+    channel_list: int = 0,
+    img_dims: str = "TYX",
+    label_dims: str = "ZYX",
+):
+    image_list = os.listdir(image_directory)
+
+    apoc.erase_classifier(cl_filename)
+    custom_classifier = apoc.PixelClassifier(opencl_filename=cl_filename)
+
+    for file in image_list:
+
+        image_stack = []
+        img = AICSImage(image_directory / file)
+
+        def channel_image(img, dims: str, channel: str or int):
+            if isinstance(channel, str):
+                channel_index = img.channel_names.index(channel)
+            elif isinstance(channel, int):
+                channel_index = channel
+            channel_img = img.get_image_data(dims, C=channel_index)
+            return channel_img
+
+        for channels in channel_list:
+            ch_img = channel_image(img=img, dims=img_dims, channel=channels)
+            image_stack.append(ch_img)
+
+        dask_stack = da.stack(image_stack, axis=0)
+
+        lbl = AICSImage(label_directory / file)
+        labels = channel_image(img=lbl, dims=label_dims, channel=0)
+
+        if predefined_features.value == 1:
+            print("custom")
+            feature_set = custom_features
+
+        else:
+            print("predefined")
+            feature_set = apoc.PredefinedFeatureSet[
+                predefined_features.name
+            ].value
+
+        custom_classifier.train(
+            features=feature_set,
+            image=dask_stack,
+            ground_truth=labels,
+            continue_training=True,
+        )
+
+    feature_importances = custom_classifier.feature_importances()
+    print("success")
+    # return pd.Series(feature_importances).plot.bar()
+
+    return feature_importances
+
+
+@magic_factory(
+    auto_call=False,
+    call_button="Batch Predict",
+    image_directory=dict(widget_type="FileEdit", mode="d"),
+    result_directory=dict(widget_type="FileEdit", mode="d"),
+    classifier_path=dict(widget_type="FileEdit", mode="r"),
+    channel_list=dict(widget_type="Select", choices=channel_nums),
+)
+def batch_predict(
+    image_directory=pathlib.Path(),
+    result_directory=pathlib.Path(),
+    classifier_path=pathlib.Path(),
+    channel_list: int = 0,
+    img_dims: str = "TYX",
+):
+    image_list = os.listdir(image_directory)
+    custom_classifier = apoc.PixelClassifier(opencl_filename=classifier_path)
+
+    for file in image_list:
+        # print('started predicting: ', file)
+        image_stack = []
+        img = AICSImage(image_directory / file)
+
+        def channel_image(img, dims: str, channel: str or int):
+            if isinstance(channel, str):
+                channel_index = img.channel_names.index(channel)
+            elif isinstance(channel, int):
+                channel_index = channel
+            channel_img = img.get_image_data(dims, C=channel_index)
+            return channel_img
+
+        for channels in channel_list:
+            ch_img = channel_image(img=img, dims=img_dims, channel=channels)
+            image_stack.append(ch_img)
+
+        dask_stack = da.stack(image_stack, axis=0)
+
+        result = custom_classifier.predict(
+            image=dask_stack,
+        )
+
+        AICSImage(cle.pull(result)).save(uri=result_directory / file)
+
+    return result

src/napari_ndev/napari.yaml

@@ -4,8 +4,18 @@ contributions:
   commands:
     - id: napari-ndev.make_batch_annotator
       python_name: napari_ndev._widget:batch_annotator
-      title: Make Batch Annotator
+      title: Make Batch Annotator Widget
+    - id: napari-ndev.make_batch_training
+      python_name: napari_ndev._widget:batch_training
+      title: Make Batch Training Widget
+    - id: napari-ndev.make_batch_predict
+      python_name: napari_ndev._widget:batch_predict
+      title: Make Batch Predict Widget
 
   widgets:
     - command: napari-ndev.make_batch_annotator
       display_name: Batch Annotator
+    - command: napari-ndev.make_batch_training
+      display_name: Batch APOC Training
+    - command: napari-ndev.make_batch_predict
+      display_name: Batch APOC Predict