Merge pull request #270 from mdekstrand/feature/split-hpf

Split HPF into a separate package

doc/addons.rst

@@ -0,0 +1,10 @@
+Add-On Algorithms
+=================
+
+Several add-on packages to LensKit provide additional collections of algorithms or bridges
+to implementations in other packages.
+
+.. toctree::
+    :caption: External Algorithms
+
+    Poisson factorization <https://lkpy.lenskit.org/projects/lenskit-hpf/>

doc/algorithms.rst

@@ -50,11 +50,13 @@ TensorFlow
     tf.IntegratedBiasMF
     tf.BPR
 
-External Library Wrappers
-~~~~~~~~~~~~~~~~~~~~~~~~~
+Add-On Packages
+~~~~~~~~~~~~~~~
+
+See `add-on algorithms <addons.rst>`_ for additional algorithm families and bridges to other
+packages.
 
 .. autosummary::
-    
+
     implicit.BPR
     implicit.ALS
-    hpf.HPF

doc/index.rst

@@ -35,7 +35,7 @@ Resources
 .. toctree::
    :maxdepth: 2
    :caption: Running Experiments
-   
+
    datasets
    crossfold
    batch
@@ -52,8 +52,8 @@ Resources
     bias
     knn
     mf
+    addons
     tf
-    hpf
     implicit
 
 .. toctree::

lenskit/algorithms/hpf.py

@@ -1,56 +1 @@
-import logging
-
-import pandas as pd
-
-from .mf_common import MFPredictor
-
-_logger = logging.getLogger(__name__)
-
-
-class HPF(MFPredictor):
-    """
-    Hierarchical Poisson factorization, provided by
-    `hpfrec <https://hpfrec.readthedocs.io/en/latest/>`_.
-
-    Args:
-        features(int): the number of features
-        **kwargs: arguments passed to :py:class:`hpfrec.HPF`.
-    """
-
-    def __init__(self, features, **kwargs):
-        self.features = features
-        self._kwargs = kwargs
-
-    def fit(self, ratings, **kwargs):
-        import hpfrec
-
-        users = pd.Index(ratings.user.unique())
-        items = pd.Index(ratings.item.unique())
-
-        if 'rating' in ratings.columns:
-            count = ratings.rating.values.copy()
-        else:
-            _logger.info('no ratings found, assuming 1.0')
-            count = 1.0
-
-        hpfdf = pd.DataFrame({
-            'UserId': users.get_indexer(ratings.user),
-            'ItemId': items.get_indexer(ratings.item),
-            'Count': count
-        })
-
-        hpf = hpfrec.HPF(self.features, reindex=False, **self._kwargs)
-
-        _logger.info('fitting HPF model with %d features', self.features)
-        hpf.fit(hpfdf)
-
-        self.user_index_ = users
-        self.item_index_ = items
-        self.user_features_ = hpf.Theta
-        self.item_features_ = hpf.Beta
-
-        return self
-
-    def predict_for_user(self, user, items, ratings=None):
-        # look up user index
-        return self.score_by_ids(user, items)
+from lenskit_hpf import HPF  # noqa: F401

lkbuild/tasks.py

@@ -3,8 +3,10 @@
 """
 
 import sys
+from pathlib import Path
 from invoke import task
 from . import env
+import yaml
 
 __ALL__ = [
     'dev_lock',
@@ -39,6 +41,39 @@ def dev_lock(c, platform=None, extras=None, version=None, blas=None, env_file=Fa
     c.run(cmd)
 
 
+@task(iterable=['extras'])
+def env_file(c, platform=None, extras=None, version=None, blas=None, dev_deps=True,
+             output=None, name=None):
+    "Create an unresolved environment file"
+    from conda_lock.conda_lock import parse_source_files, aggregate_lock_specs
+
+    if not platform:
+        platform = env.conda_platform()
+
+    files = [Path('pyproject.toml')]
+    if version:
+        files.append(Path(f'lkbuild/python-{version}-spec.yml'))
+    if blas:
+        files.append(Path(f'lkbuild/{blas}-spec.yml'))
+
+    lock = parse_source_files(files, platform, dev_deps, extras)
+    lock = aggregate_lock_specs(lock)
+    env_spec = {
+        'channels': lock.channels,
+        'dependencies': lock.specs,
+    }
+    if name:
+        env_spec['name'] = name
+
+    if output:
+        print('writing environment to', output, file=sys.stderr)
+        out = Path(output)
+        with out.open('w') as f:
+            yaml.dump(env_spec, f)
+    else:
+        yaml.dump(env_spec, sys.stdout)
+
+
 @task
 def conda_platform(c, gh_output=False):
     plat = env.conda_platform()

pyproject.toml

@@ -18,6 +18,7 @@ classifiers = [
 ]
 requires-python = ">= 3.7"
 description-file = "README.md"
+license = "MIT"
 requires = [
     "pandas >=1.0, ==1.*",
     "numpy >= 1.17",
@@ -65,7 +66,7 @@ demo = [
     "nbval >= 0.9",
     "matplotlib ~= 3.4",
 ]
-hpf = ["hpfrec"]
+hpf = ["lenskit-hpf"]
 implicit = ["implicit"]
 sklearn = ["scikit-learn >= 0.22"]
 tf = ["tensorflow >=2.1,<2.6"]