Merge pull request #128 from mdekstrand/feature/fit-kwargs

Make the fit signature consistent

doc/releases.md

@@ -4,11 +4,19 @@
 
 See the [GitHub milestone](https://github.com/lenskit/lkpy/milestone/5) for full change list.
 
+### Infrastructure Updates
+
+- Dropped support for Python 3.5
+- Removed `*args` from `Algorithm.fit`, so additional data must be provided via keyword arguments
+- Made `Algorithm.fit` implementations consistently take `**kwargs` for hybrid flexibility
+
 ### Algorithm Updates
 
+- Substantial performance and stability improvements to item-item
 - Added a coordinate descent solver to explicit-feedback ALS and made it the default.  The old
   LU-based solver is still available with `method='lu'`.
 - Added a conjugate gradient solver to implicit-feedback ALS and made it the default.
+- Added a random recommender
 
 ## 0.7.0
 

lenskit/algorithms/__init__.py

@@ -21,14 +21,15 @@ class Algorithm(metaclass=ABCMeta):
     """
 
     @abstractmethod
-    def fit(self, ratings, *args, **kwargs):
+    def fit(self, ratings, **kwargs):
         """
         Train a model using the specified ratings (or similar) data.
 
         Args:
             ratings(pandas.DataFrame): The ratings data.
-            args: Additional training data the algorithm may require.
-            kwargs: Additional training data the algorithm may require.
+            kwargs: Additional training data the algorithm may require.  Algorithms should
+                avoid using the same keyword arguments for different purposes, so that
+                they can be more easily hybridized.
 
         Returns:
             The algorithm object.

lenskit/algorithms/als.py

@@ -302,7 +302,7 @@ def __init__(self, features, *, iterations=20, reg=0.1, damping=5, bias=True, me
         self.progress = progress if progress is not None else util.no_progress
         self._random = rand
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         """
         Run ALS to train a model.
 
@@ -488,7 +488,7 @@ def __init__(self, features, *, iterations=20, reg=0.1, weight=40, method='cg',
         self._random = rand
         self.progress = progress if progress is not None else util.no_progress
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         self.timer = util.Stopwatch()
         current, uctx, ictx = self._initial_model(ratings)
 

lenskit/algorithms/basic.py

@@ -69,22 +69,22 @@ def __init__(self, items=True, users=True, damping=0.0):
         check.check_value(self.item_damping >= 0, "item damping value {} must be nonnegative",
                           self.item_damping)
 
-    def fit(self, data):
+    def fit(self, ratings, **kwargs):
         """
         Train the bias model on some rating data.
 
         Args:
-            data (DataFrame): a data frame of ratings. Must have at least `user`,
-                              `item`, and `rating` columns.
+            ratings (DataFrame): a data frame of ratings. Must have at least `user`,
+                                 `item`, and `rating` columns.
 
         Returns:
             Bias: the fit bias object.
         """
 
-        _logger.info('building bias model for %d ratings', len(data))
-        self.mean_ = data.rating.mean()
+        _logger.info('building bias model for %d ratings', len(ratings))
+        self.mean_ = ratings.rating.mean()
         _logger.info('global mean: %.3f', self.mean_)
-        nrates = data.assign(rating=lambda df: df.rating - self.mean_)
+        nrates = ratings.assign(rating=lambda df: df.rating - self.mean_)
 
         if self.items:
             group = nrates.groupby('item').rating
@@ -167,7 +167,7 @@ def __init__(self, selector=None):
         else:
             self.selector = selector
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         pop = ratings.groupby('item').user.count()
         pop.name = 'score'
         self.item_pop_ = pop.astype('float64')
@@ -236,9 +236,9 @@ def __init__(self, algorithms, *others):
         else:
             self.algorithms = [algorithms]
 
-    def fit(self, ratings, *args, **kwargs):
+    def fit(self, ratings, **kwargs):
         for algo in self.algorithms:
-            algo.fit(ratings, *args, **kwargs)
+            algo.fit(ratings, **kwargs)
 
         return self
 
@@ -285,7 +285,7 @@ def __init__(self, predictor, selector=None):
         self.predictor = predictor
         self.selector = selector if selector is not None else UnratedItemCandidateSelector()
 
-    def fit(self, ratings, *args, **kwargs):
+    def fit(self, ratings, **kwargs):
         """
         Fit the recommender.
 
@@ -296,8 +296,8 @@ def fit(self, ratings, *args, **kwargs):
             args, kwargs:
                 Additional arguments for the predictor to use in its training process.
         """
-        self.predictor.fit(ratings, *args, **kwargs)
-        self.selector.fit(ratings)
+        self.predictor.fit(ratings, **kwargs)
+        self.selector.fit(ratings, **kwargs)
         return self
 
     def recommend(self, user, n=None, candidates=None, ratings=None):
@@ -339,7 +339,7 @@ class UnratedItemCandidateSelector(CandidateSelector):
     users_ = None
     user_items_ = None
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         r2 = ratings[['user', 'item']]
         sparse = sparse_ratings(r2)
         _logger.info('trained unrated candidate selector for %d ratings', sparse.matrix.nnz)
@@ -388,8 +388,8 @@ def __init__(self, selector=None, random_state=None):
         self.random_state = random_state
         self.items = None
 
-    def fit(self, ratings, *args, **kwargs):
-        self.selector.fit(ratings)
+    def fit(self, ratings, **kwargs):
+        self.selector.fit(ratings, **kwargs)
         items = pd.DataFrame(ratings['item'].unique(), columns=['item'])
         self.items = items
         return self

lenskit/algorithms/funksvd.py

@@ -221,7 +221,7 @@ def __init__(self, features, iterations=100, *, lrate=0.001, reg=0.015,
         else:
             self.bias = bias
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         """
         Train a FunkSVD model.
 

lenskit/algorithms/hpf.py

@@ -22,7 +22,7 @@ def __init__(self, features, **kwargs):
         self.features = features
         self._kwargs = kwargs
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         import hpfrec
 
         users = pd.Index(ratings.user.unique())

lenskit/algorithms/implicit.py

@@ -31,7 +31,7 @@ class BaseRec(Recommender, Predictor):
     def __init__(self, delegate):
         self.delegate = delegate
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         matrix, users, items = sparse_ratings(ratings, scipy=True)
         iur = matrix.T.tocsr()
 

lenskit/algorithms/item_knn.py

@@ -303,7 +303,7 @@ def __init__(self, nnbrs, min_nbrs=1, min_sim=1.0e-6, save_nbrs=None,
         self.center = center
         self.aggregate = aggregate
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         """
         Train a model.
 

lenskit/algorithms/user_knn.py

@@ -124,7 +124,7 @@ def __init__(self, nnbrs, min_nbrs=1, min_sim=0, center=True, aggregate='weighte
         self.center = center
         self.aggregate = intern(aggregate)
 
-    def fit(self, ratings):
+    def fit(self, ratings, **kwargs):
         """
         "Train" a user-user CF model.  This memorizes the rating data in a format that is usable
         for future computations.