dynamic module loading

pyterrier/__init__.py

@@ -48,21 +48,6 @@
 cast = deprecated(version='0.11.0', reason="use pt.java.cast(...) instead")(java.cast)
 
 
-# Additional setup performed in a function to avoid polluting the namespace with other imports like platform
-def _():
-    # check python version
-    import platform
-    from packaging.version import Version
-    if Version(platform.python_version()) < Version('3.7.0'):
-        raise RuntimeError("From PyTerrier 0.8, Python 3.7 minimum is required, you currently have %s" % platform.python_version())
-
-    # apply is an object, not a module, as it also has __get_attr__() implemented
-    from pyterrier.apply import _apply
-    globals()['apply'] = _apply()
-
-    utils.set_tqdm()
-_()
-
 __all__ = [
     'java', 'terrier', 'anserini', 'cache', 'debug', 'io', 'measures', 'model', 'new', 'ltr', 'parallel', 'pipelines',
     'text', 'transformer', 'datasets', 'get_dataset', 'find_datasets', 'list_datasets', 'Experiment', 'GridScan',
@@ -71,9 +56,44 @@ def _():
     'BatchRetrieve', 'TerrierRetrieve', 'FeaturesBatchRetrieve', 'IndexFactory',
     'run', 'rewrite', 'index', 'FilesIndexer', 'TRECCollectionIndexer', 'DFIndexer', 'DFIndexUtils', 'IterDictIndexer',
     'IndexingType', 'TerrierStemmer', 'TerrierStopwords', 'TerrierTokeniser',
-    'IndexRef', 'ApplicationSetup', 'properties', 'apply',
+    'IndexRef', 'ApplicationSetup', 'properties',
 
     # Deprecated:
     'init', 'started', 'logging', 'version', 'check_version', 'extend_classpath', 'set_tqdm', 'set_property', 'set_properties',
     'redirect_stdouterr', 'autoclass', 'cast',
+
+    # Entry point modules (appended loaded below):
 ]
+
+
+# Additional setup performed in a function to avoid polluting the namespace with other imports like platform
+def _():
+    from warnings import warn
+    import platform
+    from packaging.version import Version
+
+    # check python version
+    if Version(platform.python_version()) < Version('3.7.0'):
+        raise RuntimeError("From PyTerrier 0.8, Python 3.7 minimum is required, you currently have %s" % platform.python_version())
+
+    globs = globals()
+
+    # Load the _apply object as pt.apply so that the dynamic __getattr__ methods work
+    from pyterrier.apply import _apply
+    globs['apply'] = _apply()
+    __all__.append('apply')
+
+    # load modules defined as package entry points into the global pyterrier namespace
+    for entry_point in utils.entry_points('pyterrier.modules'):
+        if entry_point.name in globs:
+            warn(f'skipping loading {entry_point} because a module with this name is already loaded.')
+            continue
+        module = entry_point.load()
+        if callable(module): # if the entry point refers to an function/class, call it to get the module
+            module = module()
+        globs[entry_point.name] = module
+        __all__.append(entry_point.name)
+
+    # guess the environment and set an appropriate tqdm as pt.tqdm
+    utils.set_tqdm()
+_()

pyterrier/ltr.py

@@ -1,6 +1,5 @@
-
+import pyterrier as pt
 from . import Transformer, Estimator
-from .apply import doc_score, doc_features
 from .model import add_ranks
 from typing import Sequence, Union, Tuple
 import numpy as np, pandas as pd
@@ -244,7 +243,7 @@ def feature_to_score(fid : int) -> Transformer:
         Args: 
             fid: a single feature id that should be kept
     """
-    return doc_score(lambda row : row["features"][fid])
+    return pt.apply.doc_score(lambda row : row["features"][fid])
 
 def apply_learned_model(learner, form : str = 'regression', **kwargs) -> Transformer:
     """
@@ -284,4 +283,4 @@ def score_to_feature() -> Transformer:
             three_features = cands >> (bm25f  **  pl2f ** pt.ltr.score_to_feature())  
 
     """
-    return doc_features(lambda row : np.array(row["score"]))
+    return pt.apply.doc_features(lambda row : np.array(row["score"]))