Add auto to umap (#197)

* adds auto to `init_pos` * add auto to umap * adds release notes * adds note * update note * update note * make nicer docstring
scverse · May 17, 2024 · a131b72 · a131b72
1 parent 10d52d5
commit a131b72
Show file tree

Hide file tree

Showing 4 changed files with 32 additions and 5 deletions.
diff --git a/docs/release-notes/0.10.4.md b/docs/release-notes/0.10.4.md
@@ -1,4 +1,4 @@
-### 0.10.3 {small}`2024-05-14`
+### 0.10.4 {small}`2024-05-14`
 
 ```{rubric} Features
 ```

diff --git a/docs/release-notes/0.10.5.md b/docs/release-notes/0.10.5.md
@@ -0,0 +1,15 @@
+### 0.10.5 {small}`the future`
+
+```{rubric} Features
+```
+
+```{rubric} Performance
+```
+
+```{rubric} Bug fixes
+```
+* fixes an issue with `tl.umap` and `init_pos` {pr}`193` {smaller}`S Dicks`
+
+
+```{rubric} Misc
+```
diff --git a/docs/release-notes/index.md b/docs/release-notes/index.md
@@ -3,6 +3,8 @@
 # Release notes
 
 ## Version 0.10.0
+```{include} /release-notes/0.10.5.md
+```
 ```{include} /release-notes/0.10.4.md
 ```
 ```{include} /release-notes/0.10.3.md

diff --git a/src/rapids_singlecell/tools/_umap.py b/src/rapids_singlecell/tools/_umap.py
@@ -15,7 +15,7 @@
 if TYPE_CHECKING:
     from anndata import AnnData
 
-_InitPos = Literal["spectral", "random"]
+_InitPos = Literal["auto", "spectral", "random"]
 
 
 def umap(
@@ -27,7 +27,7 @@ def umap(
     maxiter: int | None = None,
     alpha: float = 1.0,
     negative_sample_rate: int = 5,
-    init_pos: _InitPos = "spectral",
+    init_pos: _InitPos = "auto",
     random_state=0,
     a: float | None = None,
     b: float | None = None,
@@ -72,8 +72,14 @@ def umap(
     init_pos
         How to initialize the low dimensional embedding. Called `init` in the
         original UMAP. Options are:
-        * 'spectral': use a spectral embedding of the graph.
-        * 'random': assign initial embedding positions at random.
+
+            * 'auto': chooses 'spectral' for `'n_samples' < 1000000`, 'random' otherwise.
+            * 'spectral': use a spectral embedding of the graph.
+            * 'random': assign initial embedding positions at random.
+
+        .. note::
+            If your embedding looks odd it's recommended setting `init_pos` to 'random'.
+
     random_state
         `int`, `random_state` is the seed used by the random number generator
     a
@@ -150,6 +156,10 @@ def umap(
         pre_knn = (knn_indices, knn_dist)
     else:
         pre_knn = None
+
+    if init_pos == "auto":
+        init_pos = "spectral" if n_obs < 1000000 else "random"
+
     umap = UMAP(
         n_neighbors=n_neighbors,
         n_components=n_components,