adding histnet

docs/source/quapy.method.rst

@@ -17,7 +17,7 @@ quapy.method.aggregative module
    :undoc-members:
    :show-inheritance:
 
-.. automodule:: quapy.method._neural
+.. automodule:: quapy.method._quanet
    :members:
    :undoc-members:
    :show-inheritance:

quapy/__init__.py

@@ -17,7 +17,7 @@ try:
 except ImportError:
     plot = None
 
-__version__ = '0.2.1.post1'
+__version__ = '0.2.2'
 
 
 def _default_cls():

quapy/method/_histnet.py

@@ -1,17 +1,21 @@
 """
-HistNetQ implementation.
+HistNetQ implementation, see the original paper:
+
+Pérez-Mon, O., Moreo, A., Coz, JJ del, & González, P. (2025).
+Quantification using permutation-invariant networks based on histograms.
+Neural Computing and Applications, 37(5), 3505-3520.
 
 Ported from the reference implementation at https://github.com/pglez84/histnetq (the `HistNet`/
 `DLQuantification` classes in that repo), adapted to QuaPy's own protocol-based sample generation
 (replacing that repo's custom, `quantificationlib`-backed bag generators) and restricted, for now, to
-the "hard" differentiable histogram variant:
+the "hard" differentiable histogram variant by:
 
 Yusuf, I., Igwegbe, G., and Azeez, O. "Differentiable Histogram with Hard-Binning."
 arXiv preprint arXiv:2012.06311 (2020).
 
 The overall architecture is: feature_extraction -> Sigmoid -> histogram layer -> small MLP -> softmax,
 trained by minimizing a quantification loss over samples ("bags") of known prevalence, rather than
-over individually labeled instances (in the spirit of QuaNet, see method/_neural.py).
+over individually labeled instances (in the spirit of QuaNet, see method/_quanet.py).
 """
 import copy
 import os
@@ -177,10 +181,17 @@ class HistNetQ(BaseQuantifier):
     Implementation of `HistNetQ <https://github.com/pglez84/histnetq>`_, a neural network for
     quantification that learns a differentiable histogram-based representation of a sample, trained
     end-to-end by minimizing a quantification loss over many samples ("bags") of known prevalence.
+    The method was proposed in `Pérez-Mon, O., Moreo, A., Coz, JJ del, & González, P. (2025).
+    Quantification using permutation-invariant networks based on histograms.
+    Neural Computing and Applications, 37(5), 3505-3520.
+    <https://link.springer.com/article/10.1007/s00521-024-10721-1>`_
 
     HistNetQ does not follow the classify-then-aggregate pattern of :class:`quapy.method.aggregative.
-    AggregativeQuantifier`; like :class:`quapy.method.meta.QuaNet`, it is trained and evaluated
+    AggregativeQuantifier`. Such classical approach is termed asymmetric, in the sense that quantifiers
-    end-to-end on whole samples rather than on individually labeled instances.
+    learn from labelled instances and perform inference over bags.
+    Like :class:`quapy.method.meta.QuaNet`, HistNetQ is trained and evaluated
+    end-to-end on whole samples rather than on individually labeled instances, following a symmetric problem setting
+    (learning from bags, predicting on bags).
 
     Training data can be provided in two ways:
 

quapy/method/meta.py

@@ -16,13 +16,13 @@ from quapy.method.base import BaseQuantifier, BinaryQuantifier
 from quapy.method.aggregative import CC, ACC, PACC, HDy, EMQ, AggregativeQuantifier, AggregativeSoftQuantifier
 
 try:
-    from . import _neural
+    from . import _quanet
 except ModuleNotFoundError:
-    _neural = None
+    _quanet = None
 
 
-if _neural:
+if _quanet:
-    QuaNet = _neural.QuaNetTrainer
+    QuaNet = _quanet.QuaNetTrainer
 else:
     QuaNet = "QuaNet is not available due to missing torch package"
 

quapy/tests/test_methods.py

@@ -126,6 +126,7 @@ class TestMethods(unittest.TestCase):
             estim_prevalences2 = model2.predict(dataset.test.X)
             self.assertTrue(check_prevalence_vector(estim_prevalences2))
 
+
     def test_composable(self):
         try:
             from quapy.method.composable import check_compatible_qunfold_version