update changelog and svm au model note in docs

docs/pages/changelog.md

@@ -25,6 +25,7 @@ This is a large overhaul and refactor of some of the core testing and API functi
 - the new default model for AU detection was changed to our new `xgb` model which gives continuous valued predictions between 0-1
 - remove support for `fer` emotion model
 - remove support for `jaanet` AU model
+- remove support for `logistic` AU model
 - remove support for `pnp` facepose detector
 - drop support for reading and manipulating Affectiva and FACET data
 - `.detect_image` will no longer resize images on load as the new default for `output_size=None`. If you want to process images with `batch_size > 1` and images differ in size, then you will be **required** to manually set `output_size` otherwise py-feat will raise a helpful error message

docs/pages/models.md

@@ -32,12 +32,16 @@ Models names are case-insensitive: `'resmasknet' == 'ResMaskNet'`
 
 ## Action Unit detection
 - **`xgb`: XGBoost Classifier model trained on Histogram of Oriented Gradients\*** extracted from BP4D, DISFA, CK+, UNBC-McMaster shoulder pain, and AFF-Wild2 datasets  
-- `svm`: SVM model trained on Histogram of Oriented Gradients extracted from BP4D, DISFA, CK+, UNBC-McMaster shoulder pain, and AFF-Wild2 datasets
+- `svm`: SVM model trained on Histogram of Oriented Gradients\*\* extracted from BP4D, DISFA, CK+, UNBC-McMaster shoulder pain, and AFF-Wild2 datasets
 
 ```{note}
 \*For AU07, our `xbg` detector was trained with hinge-loss instead of cross-entropy loss like other AUs as this yielded substantially better detection peformance given the labeled data available for this AU. This means that while it returns continuous probability predictions,  these are more likely to appear binary in practice (i.e. be 0 or 1) and should be interpreted as *proportion of decision-trees with a detection* rather than *average decision-tree confidence* like other AU values.
 ```
 
+```{note}
+\*\* Our `svm` detector uses the [`LinearSVC`](https://scikit-learn.org/stable/modules/generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC) implementation from `sklearn` and thus returns **binary values** for each AU rather than probabilities. If your use-case requires continuous-valued detections, we recommend the `xgb` detector instead. 
+```
+
 ## Emotion detection
 - **`resmasknet`: Facial expression recognition using residual masking network** by ([Pham et al., 2020](https://ieeexplore.ieee.org/document/9411919))
 - `svm`: SVM model trained on Histogram of Oriented Gradients extracted from ExpW, CK+, and JAFFE datasets