Release 0.6.4 (#76)

* update contact email

* filter values for flesch and sentence length

* Release 0.6.3 (#73) (#74)

* update contact email

* use alternative version of textstat for flesch scores

* filter values for flesch and sentence length

* add argument to return single prediction for multiclass probability estimate (#75)

* Update version 0.6.4

quantgov/__init__.py

@@ -4,4 +4,4 @@
 from . import corpus, nlp, ml, utils
 from .utils import load_driver
 
-__version__ = '0.6.3'
+__version__ = '0.6.4'

quantgov/__main__.py

@@ -128,6 +128,9 @@ def parse_args():
     estimate.add_argument(
         '--precision', default=4, type=int,
         help='number of decimal places to round the probabilities')
+    estimate.add_argument(
+        '--oneclass', action='store_true',
+        help='only return predicted class for multiclass probabilty estimates')
     estimate.add_argument(
         '-o', '--outfile',
         type=lambda x: open(x, 'w', newline='', encoding='utf-8'),
@@ -223,7 +226,8 @@ def run_estimator(args):
                 args.estimator,
                 args.corpus,
                 args.probability,
-                args.precision)
+                args.precision,
+                args.oneclass)
         )
 
 

quantgov/ml/estimation.py

@@ -4,6 +4,7 @@
 Functionality for making predictions with an estimator
 """
 import logging
+import numpy as np
 
 log = logging.getLogger(__name__)
 
@@ -106,7 +107,7 @@ def estimate_probability_multilabel(estimator, streamer, precision):
         )
 
 
-def estimate_probability_multiclass(estimator, streamer, precision):
+def estimate_probability_multiclass(estimator, streamer, precision, oneclass):
     """
     Generate probabilities for a one-label, multiclass estimator
 
@@ -119,12 +120,24 @@ def estimate_probability_multiclass(estimator, streamer, precision):
 
     """
     texts = (doc.text for doc in streamer)
-    probs = estimator.pipeline.predict_proba(texts).round(precision)
-    yield from (
-        (docidx, (class_, probability))
-        for docidx, doc_probs in zip(streamer.index, probs)
-        for class_, probability in zip(estimator.pipeline.classes_, doc_probs)
-    )
+    probs = estimator.pipeline.predict_proba(texts)
+    # If oneclass flag is true, only returns the predicted class
+    if oneclass:
+        class_indices = list(i[-1] for i in np.argsort(probs, axis=1))
+        yield from (
+            (docidx, (estimator.pipeline.classes_[class_index],
+                      doc_probs[class_index].round(precision)))
+            for docidx, doc_probs, class_index in zip(
+                streamer.index, probs, class_indices)
+        )
+    # Else returns probabilty values for all classes
+    else:
+        yield from (
+            (docidx, (class_, probability.round(precision)))
+            for docidx, doc_probs in zip(streamer.index, probs)
+            for class_, probability in zip(
+                estimator.pipeline.classes_, doc_probs)
+        )
 
 
 def estimate_probability_multilabel_multiclass(estimator, streamer, precision):
@@ -140,7 +153,7 @@ def estimate_probability_multilabel_multiclass(estimator, streamer, precision):
 
     """
     texts = (doc.text for doc in streamer)
-    probs = estimator.pipeline.predict_proba(texts)
+    probs = estimator.pipeline.predict_proba(texts).round(precision)
     yield from (
         (docidx, (label_name, class_, prob))
         for label_name, label_probs in zip(estimator.label_names, probs)
@@ -149,7 +162,8 @@ def estimate_probability_multilabel_multiclass(estimator, streamer, precision):
     )
 
 
-def estimate(estimator, corpus, probability, precision=4, *args, **kwargs):
+def estimate(estimator, corpus, probability, precision=4, oneclass=False,
+             *args, **kwargs):
     """
     Estimate label values for documents in corpus
 
@@ -171,7 +185,7 @@ def estimate(estimator, corpus, probability, precision=4, *args, **kwargs):
                     estimator, streamer, precision)
         elif estimator.multiclass:  # Multiclass probability
             yield from estimate_probability_multiclass(
-                estimator, streamer, precision)
+                estimator, streamer, precision, oneclass)
         else:  # Simple probability
             yield from estimate_probability(
                 estimator, streamer, precision)

tests/test_ml.py

@@ -65,3 +65,14 @@ def test_multiclass_probability_estimator():
                       'moby,money,0.1536\n'
                       'moby,science-and-technology,0.1671\n'
                       'moby,world,0.141\n')
+
+
+def test_multiclass_probability_oneclass_estimator():
+    output = check_output(
+        ['quantgov', 'ml', 'estimate',
+         str(PSEUDO_ESTIMATOR_PATH.joinpath('data', 'multiclass.qge')),
+         str(PSEUDO_CORPUS_PATH), '--probability', '--oneclass']
+    )
+    assert output == ('file,class,probability\n'
+                      'cfr,world,0.1997\n'
+                      'moby,health-and-public-welfare,0.205\n')