added unsupervised ml exercise

ScaDS · Nov 26, 2023 · 3419139 · 3419139
1 parent f04a49c
commit 3419139
Show file tree

Hide file tree

Showing 41 changed files with 117 additions and 8 deletions.
diff --git a/...eractive_object_classification/apoc21.png → ...eractive_object_classification/apoc21.png b/...eractive_object_classification/apoc21.png → ...eractive_object_classification/apoc21.png
diff --git a/...eractive_object_classification/apoc22.png → ...eractive_object_classification/apoc22.png b/...eractive_object_classification/apoc22.png → ...eractive_object_classification/apoc22.png
diff --git a/...eractive_object_classification/apoc23.png → ...eractive_object_classification/apoc23.png b/...eractive_object_classification/apoc23.png → ...eractive_object_classification/apoc23.png
diff --git a/...eractive_object_classification/apoc24.png → ...eractive_object_classification/apoc24.png b/...eractive_object_classification/apoc24.png → ...eractive_object_classification/apoc24.png
diff --git a/...eractive_object_classification/apoc25.png → ...eractive_object_classification/apoc25.png b/...eractive_object_classification/apoc25.png → ...eractive_object_classification/apoc25.png
diff --git a/...eractive_object_classification/apoc26.png → ...eractive_object_classification/apoc26.png b/...eractive_object_classification/apoc26.png → ...eractive_object_classification/apoc26.png
diff --git a/...eractive_object_classification/apoc27.png → ...eractive_object_classification/apoc27.png b/...eractive_object_classification/apoc27.png → ...eractive_object_classification/apoc27.png
diff --git a/...eractive_object_classification/apoc28.png → ...eractive_object_classification/apoc28.png b/...eractive_object_classification/apoc28.png → ...eractive_object_classification/apoc28.png
diff --git a/...eractive_object_classification/apoc29.png → ...eractive_object_classification/apoc29.png b/...eractive_object_classification/apoc29.png → ...eractive_object_classification/apoc29.png
diff --git a/...eractive_object_classification/apoc30.png → ...eractive_object_classification/apoc30.png b/...eractive_object_classification/apoc30.png → ...eractive_object_classification/apoc30.png
diff --git a/...eractive_object_classification/apoc31.png → ...eractive_object_classification/apoc31.png b/...eractive_object_classification/apoc31.png → ...eractive_object_classification/apoc31.png
diff --git a/...eractive_object_classification/apoc32.png → ...eractive_object_classification/apoc32.png b/...eractive_object_classification/apoc32.png → ...eractive_object_classification/apoc32.png
diff --git a/...eractive_object_classification/apoc33.png → ...eractive_object_classification/apoc33.png b/...eractive_object_classification/apoc33.png → ...eractive_object_classification/apoc33.png
diff --git a/...eractive_object_classification/apoc34.png → ...eractive_object_classification/apoc34.png b/...eractive_object_classification/apoc34.png → ...eractive_object_classification/apoc34.png
diff --git a/...sification/apoc_object_classification.mp4 → ...sification/apoc_object_classification.mp4 b/...sification/apoc_object_classification.mp4 → ...sification/apoc_object_classification.mp4
diff --git a/...teractive_object_classification/readme.md → ...teractive_object_classification/readme.md b/...teractive_object_classification/readme.md → ...teractive_object_classification/readme.md
@@ -24,7 +24,8 @@ or using the menu `Tools > Segmentation / labeling > Gauss-Otsu Labeling (clespe
 
 ## Object classification
 
-Our starting point is a loaded image and a label image with segmented objects. The following procedure is also shown in [this video](apoc_object_classification.mp4).
+Our starting point is a loaded image and a label image with segmented objects. 
+The following procedure is also shown in [this video](apoc_object_classification.mp4).
 
 ![](apoc21.png)
 

diff --git a/...nteractive_pixel_classification/apoc1.png → ...nteractive_pixel_classification/apoc1.png b/...nteractive_pixel_classification/apoc1.png → ...nteractive_pixel_classification/apoc1.png
diff --git a/...teractive_pixel_classification/apoc10.png → ...teractive_pixel_classification/apoc10.png b/...teractive_pixel_classification/apoc10.png → ...teractive_pixel_classification/apoc10.png
diff --git a/...teractive_pixel_classification/apoc11.png → ...teractive_pixel_classification/apoc11.png b/...teractive_pixel_classification/apoc11.png → ...teractive_pixel_classification/apoc11.png
diff --git a/...nteractive_pixel_classification/apoc2.png → ...nteractive_pixel_classification/apoc2.png b/...nteractive_pixel_classification/apoc2.png → ...nteractive_pixel_classification/apoc2.png
diff --git a/...nteractive_pixel_classification/apoc3.png → ...nteractive_pixel_classification/apoc3.png b/...nteractive_pixel_classification/apoc3.png → ...nteractive_pixel_classification/apoc3.png
diff --git a/...nteractive_pixel_classification/apoc4.png → ...nteractive_pixel_classification/apoc4.png b/...nteractive_pixel_classification/apoc4.png → ...nteractive_pixel_classification/apoc4.png
diff --git a/...nteractive_pixel_classification/apoc5.png → ...nteractive_pixel_classification/apoc5.png b/...nteractive_pixel_classification/apoc5.png → ...nteractive_pixel_classification/apoc5.png
diff --git a/...nteractive_pixel_classification/apoc6.png → ...nteractive_pixel_classification/apoc6.png b/...nteractive_pixel_classification/apoc6.png → ...nteractive_pixel_classification/apoc6.png
diff --git a/...nteractive_pixel_classification/apoc7.png → ...nteractive_pixel_classification/apoc7.png b/...nteractive_pixel_classification/apoc7.png → ...nteractive_pixel_classification/apoc7.png
diff --git a/...nteractive_pixel_classification/apoc8.png → ...nteractive_pixel_classification/apoc8.png b/...nteractive_pixel_classification/apoc8.png → ...nteractive_pixel_classification/apoc8.png
diff --git a/...nteractive_pixel_classification/apoc9.png → ...nteractive_pixel_classification/apoc9.png b/...nteractive_pixel_classification/apoc9.png → ...nteractive_pixel_classification/apoc9.png
diff --git a/...assification/apoc_object_segmentation.mp4 → ...assification/apoc_object_segmentation.mp4 b/...assification/apoc_object_segmentation.mp4 → ...assification/apoc_object_segmentation.mp4
diff --git a/...nteractive_pixel_classification/readme.md → ...nteractive_pixel_classification/readme.md b/...nteractive_pixel_classification/readme.md → ...nteractive_pixel_classification/readme.md
diff --git a/04_unsupervised_machine_learning/img.png b/04_unsupervised_machine_learning/img.png
diff --git a/04_unsupervised_machine_learning/img_1.png b/04_unsupervised_machine_learning/img_1.png
diff --git a/04_unsupervised_machine_learning/img_2.png b/04_unsupervised_machine_learning/img_2.png
diff --git a/04_unsupervised_machine_learning/img_3.png b/04_unsupervised_machine_learning/img_3.png
diff --git a/04_unsupervised_machine_learning/img_4.png b/04_unsupervised_machine_learning/img_4.png
diff --git a/04_unsupervised_machine_learning/img_5.png b/04_unsupervised_machine_learning/img_5.png
diff --git a/04_unsupervised_machine_learning/img_6.png b/04_unsupervised_machine_learning/img_6.png
diff --git a/04_unsupervised_machine_learning/img_7.png b/04_unsupervised_machine_learning/img_7.png
diff --git a/04_unsupervised_machine_learning/img_8.png b/04_unsupervised_machine_learning/img_8.png
diff --git a/04_unsupervised_machine_learning/interactive_clustering.md b/04_unsupervised_machine_learning/interactive_clustering.md
@@ -0,0 +1,109 @@
+# Interactive object clustering
+
+In this exercise we will use the [napari-clusters-plotter](https://www.napari-hub.org/plugins/napari-clusters-plotter) to group objects together based on their measured properties.
+For these measurements we will use [napari-skimage-regionprops]().
+
+## Starting point
+
+Open a terminal window and activate your conda environment:
+
+```
+conda activate devbio-napari-env
+```
+
+Afterwards, start up Napari:
+
+```
+napari
+```
+
+In Napari open the "Human mitosis" example dataset from the menu `File > Open Sample > Napari builtins > Human mitosis`.
+
+![img.png](img.png)
+
+## Object segmentation
+
+Segment the nuclei using the menu `Tools > Segmentation / Labeling > Gauss-Otsu Labeling (clesperanto)`. 
+Keep the default settings and click `Run`.
+
+![img_1.png](img_1.png)
+
+Use the small hide icon to close the Gauss-Otsu-Labeling widget.
+
+## Feature extraction
+
+Measure shape and intensity features using the menu `Plugins > Measurement Tables > Regionprops (scikit-image, nsr)`.
+Make sure that the `intensity`, `size` and `shape` checkboxes are ticked and click `Run`.
+
+![img_2.png](img_2.png)
+
+Hide both, the Regionprops widget and the Table widget that just popped up.
+
+## Dimensionality reduction
+
+Generate a [UMAP](https://umap-learn.readthedocs.io/) using the menu `Plugins > Measurement post-processing > Dimensionality Reduction > UMAP (nsr)`.
+* Make sure the labels layer is selected where you just did your measurements. 
+* Choose the method `UMAP` and keep its default settings. 
+* Untick the features `bbox_area`, and `local_centroid1` / `2` using the `CTRL` key.
+* Click `Run`.
+* Wait a minute.
+
+![img_3.png](img_3.png)
+
+Close both, the Dimensionality Reduction widget and the Table widget that just popped up.
+
+## Plot measurements
+
+Open the plot widget using the menu `Plugins > Visualization > Plot measurements (ncp)`. 
+You can play a bit with columns to plot. Eventually select `UMAP_0` and `UMAP_2` as `Axes` and click on `Plot`.
+
+![img_4.png](img_4.png)
+
+
+## Manual clustering
+
+Click within the plot and before releasing the mouse button, drag the mouse to select a region of interest.
+Repeat this while holding the `CTRL` key to select multiple regions of interest.
+
+![manual_clustering.gif](manual_clustering.gif)
+
+The object selection might be related to their shape and size. 
+To visualize this hypothesis, you can select the `area` and `roundness` as Axes in the plot widget.
+Make also sure the clustering `MANUAL_CLUSTER_ID` is selected.
+Click on `Plot` again.
+
+![img_5.png](img_5.png)
+
+Close the plot widget.
+
+## Automatic clustering
+
+You can also cluster the objects automatically using the menu `Plugins > Measurement post-processing > Clustering (nsr)` menu.
+Choose the layer of the segmented and measured objects. 
+Unselect the `bbox_area` and `local_centroid1` / `2` features. 
+Unselect `aspect_ratio` because it sometimes contains `inf` values which are not supported by the clustering algorithm.
+Also unselect `UMAP_0` and `UMAP_1` as these two contain compressed information about all other columns.
+Select `K-Means` clustering and activate the `Standardize features` checkbox.
+Click on `Run`.
+
+![img_6.png](img_6.png)
+
+Close the Clustering widget and the Table widget that just popped up.
+
+## Visualizing automatic clustering
+
+To visualize the results of the automated clustering, open the plot widget again using the menu `Plugins > Visualization > Plot measurements (ncp)`.
+Select `UMAP_0` and `UMAP_1` as `Axes` and select `KMEANS_CLUSTER_ID` as `Clustering`.
+
+![img_7.png](img_7.png)
+
+## Exercise
+
+Cluster the objects into 5 classes while having only `UMAP_0` and `UMAP_1` selected in the Clustering widget. 
+Give this clustering result a name.
+Visualize the resulting clustering using the plot widget.
+The result should approximately look like this:
+
+![img_8.png](img_8.png)
+
+
diff --git a/04_unsupervised_machine_learning/manual_clustering.gif b/04_unsupervised_machine_learning/manual_clustering.gif
diff --git a/_toc.yml b/_toc.yml
@@ -11,12 +11,11 @@ parts:
     - file: 01_scripting/napari_intro
     - file: 02_notebooks/napari-assistant
     - file: 02_notebooks/notebook_export
-    - file: 03_machine_learning/interactive_pixel_classification
-    - file: 03_machine_learning/interactive_object_classification
-    # generating notebooks using APOC
-
-#    - file: 02_python_basics/readme
-#      sections:
-#      - file: 02_python_basics/01_our_first_juptyer_notebook
+
+  - caption: Optional exercises
+    chapters:
+    - file: 03_supervised_machine_learning/interactive_pixel_classification/readme
+    - file: 03_supervised_machine_learning/interactive_object_classification/readme
+    - file: 04_unsupervised_machine_learning/interactive_clustering
 
     - file: imprint