From e6f03c37c5af665c7bbcb2774fd5d3c985383c84 Mon Sep 17 00:00:00 2001
From: Mirko Bunse <mirko.bunse@cs.tu-dortmund.de>
Date: Mon, 23 Sep 2024 11:42:53 +0200
Subject: [PATCH] QuaPyWrapper supports likelihood methods

---
 qunfold/quapy.py          |  6 ++++--
 qunfold/tests/__init__.py | 31 +++++++++++++++++++++++++------
 2 files changed, 29 insertions(+), 8 deletions(-)

diff --git a/qunfold/quapy.py b/qunfold/quapy.py
index a84d42f..54ac2c1 100644
--- a/qunfold/quapy.py
+++ b/qunfold/quapy.py
@@ -62,7 +62,7 @@ class QuaPyWrapper(BaseQuantifier):
     """A thin wrapper for using qunfold methods in QuaPy.
 
     Args:
-        generic_method: A LinearMethod method to wrap.
+        generic_method: An instance of `qunfold.methods.AbstractMethod` to wrap.
 
     Examples:
         Here, we wrap an instance of ACC to perform a grid search with QuaPy.
@@ -87,4 +87,6 @@ def set_params(self, **params):
         _set_params(self.generic_method, self.get_params(deep=True), **params)
         return self
     def get_params(self, deep=True):
-        return _get_params(self.generic_method, deep, LinearMethod)
+        if isinstance(self.generic_method, LinearMethod): # use super-class constructor?
+            return _get_params(self.generic_method, deep, LinearMethod)
+        return _get_params(self.generic_method, deep)
diff --git a/qunfold/tests/__init__.py b/qunfold/tests/__init__.py
index d699d09..8c51f98 100644
--- a/qunfold/tests/__init__.py
+++ b/qunfold/tests/__init__.py
@@ -141,13 +141,18 @@ def test_methods(self):
         n_estimators = 10,
         random_state = RNG.randint(np.iinfo("uint16").max),
       )
-      p_acc = QuaPyWrapper(qunfold.ACC(lr))
+      wrapped_acc = QuaPyWrapper(qunfold.ACC(lr))
+      wrapped_sld = QuaPyWrapper(qunfold.ExpectationMaximizer(lr.estimator))
       self.assertEqual( # check that get_params returns the correct settings
-        p_acc.get_params(deep=True)["representation__classifier__estimator__C"],
+        wrapped_acc.get_params(deep=True)["representation__classifier__estimator__C"],
         1e-2
       )
-      quapy_method = qp.model_selection.GridSearchQ(
-        model = p_acc,
+      self.assertEqual(
+        wrapped_sld.get_params(deep=True)["classifier__C"],
+        1e-2
+      )
+      cv_acc = qp.model_selection.GridSearchQ(
+        model = wrapped_acc,
         param_grid = {
           "representation__classifier__estimator__C": [1e-1, 1e0, 1e1, 1e2],
         },
@@ -157,8 +162,22 @@ def test_methods(self):
         verbose = True,
       ).fit(qp.data.LabelledCollection(X_trn, y_trn))
       self.assertEqual( # check that best parameters are actually used
-        quapy_method.best_params_["representation__classifier__estimator__C"],
-        quapy_method.best_model_.generic_method.representation.classifier.estimator.C
+        cv_acc.best_params_["representation__classifier__estimator__C"],
+        cv_acc.best_model_.generic_method.representation.classifier.estimator.C
+      )
+      cv_sld = qp.model_selection.GridSearchQ(
+        model = wrapped_sld,
+        param_grid = {
+          "classifier__C": [1e-1, 1e0, 1e1, 1e2],
+        },
+        protocol = SingleSampleProtocol(X_tst, p_tst),
+        error = "mae",
+        refit = False,
+        verbose = True,
+      ).fit(qp.data.LabelledCollection(X_trn, y_trn))
+      self.assertEqual( # check that best parameters are actually used
+        cv_sld.best_params_["classifier__C"],
+        cv_sld.best_model_.generic_method.classifier.C
       )
 
 class TestDistanceRepresentation(TestCase):