Basic tutorial (#303)

Adds a simple tutorial start-to-finish tutorial for DaCapo.
Still a few things left to resolve before its ready for merging:

- [ ] Loss behaving strangely after the first validation (temporary
solution involves giving an extra singleton dimension to the provided
data)
- [ ] Validation fails (we want to show validation loss/scores,
validation outputs across iterations, and how to finally take the best
iteration and apply it with the post processing to a volume manually)
- [ ] stdout too verbose
- [ ] plotting functions should be built into `DaCapo` (there are
plotting functions, not sure how to include bokeh plots so I didn't dive
too deep into this)

.github/workflows/docs.yaml

@@ -17,7 +17,11 @@ jobs:
         fetch-depth: 0 # otherwise, you will failed to push refs to dest repo
     - name: install dacapo
       # run: pip install .[docs]
-      run: pip install sphinx-autodoc-typehints sphinx-autoapi sphinx-click sphinx-rtd-theme myst-parser
+      run: pip install sphinx-autodoc-typehints sphinx-autoapi sphinx-click sphinx-rtd-theme myst-parser jupytext ipykernel nbsphinx
+    - name: parse notebooks
+      run: jupytext --to notebook --execute ./docs/source/notebooks/*.py
+    - name: remove notebook scripts
+      run: rm ./docs/source/notebooks/*.py
     - name: Build and Commit
       uses: sphinx-notes/pages@v2
       with:

dacapo/experiments/tasks/losses/affinities_loss.py

@@ -79,10 +79,15 @@ def compute(self, prediction, target, weight):
             weight[:, self.num_affinities :, ...],
         )
 
-        return (
-            torch.nn.BCEWithLogitsLoss(reduction="none")(affs, affs_target)
-            * affs_weight
-        ).mean() + self.lsds_to_affs_weight_ratio * (
-            torch.nn.MSELoss(reduction="none")(torch.nn.Sigmoid()(aux), aux_target)
-            * aux_weight
-        ).mean()
+        if aux.shape[1] == 0:
+            return torch.nn.BCEWithLogitsLoss(reduction="none")(
+                affs, affs_target
+            ).mean()
+        else:
+            return (
+                torch.nn.BCEWithLogitsLoss(reduction="none")(affs, affs_target)
+                * affs_weight
+            ).mean() + self.lsds_to_affs_weight_ratio * (
+                torch.nn.MSELoss(reduction="none")(torch.nn.Sigmoid()(aux), aux_target)
+                * aux_weight
+            ).mean()

docs/source/index.rst

@@ -9,6 +9,7 @@
 
    overview
    install
+   notebooks/minimal_tutorial
    tutorial
    docker
    aws