improve docs

docs/biased-mf.rst

@@ -0,0 +1,22 @@
+Biased MF
+=========
+
+These models implement the standard biased matrix factorization model, like
+:py:class:`lenskit.algorithms.als.BiasedMF`, but learn the model parameters
+using TensorFlow's gradient descent instead of the alternating least squares
+algorithm.  There are two implementations:
+
+* :py:class:`lenskit_tf.BiasedMF` learns a matrix factorization to predict the
+  residuals of :py:class:`lenskit.algorithms.bias.Bias`.
+* :py:class:`lenskit_tf.IntegratedBiasMF` uses TensorFlow to learn the entire
+  model, including both biases and embeddings.
+
+Bias-Based
+~~~~~~~~~~
+
+.. autoclass:: lenskit_tf.BiasedMF
+
+Fully Integrated
+~~~~~~~~~~~~~~~~
+
+.. autoclass:: lenskit_tf.IntegratedBiasMF

docs/bpr.rst

@@ -0,0 +1,6 @@
+Bayesian Personalized Ranking
+=============================
+
+This is a TensorFlow-based implementation of BPR.
+
+.. autoclass:: BPR

docs/index.rst

@@ -28,25 +28,5 @@ Or (preferred, once published)::
 .. toctree::
     :caption: Contents
 
-Biased MF
----------
-
-These models implement the standard biased matrix factorization model, like
-:py:class:`lenskit.algorithms.als.BiasedMF`, but learn the model parameters
-using TensorFlow's gradient descent instead of the alternating least squares
-algorithm.
-
-Bias-Based
-~~~~~~~~~~
-
-.. autoclass:: BiasedMF
-
-Fully Integrated
-~~~~~~~~~~~~~~~~
-
-.. autoclass:: IntegratedBiasMF
-
-Bayesian Personalized Rating
-----------------------------
-
-.. autoclass:: BPR
+    biased-mf
+    bpr