classification: Add ModelWithThreshold

Author: janezd

Orange/classification/__init__.py

@@ -19,3 +19,4 @@
 from .rules import *
 from .sgd import *
 from .neural_network import *
+from .calibration import *

Orange/classification/calibration.py

@@ -0,0 +1,22 @@
+from Orange.classification import Model
+
+__all__ = ["ModelWithThreshold"]
+
+
+class ModelWithThreshold(Model):
+    def __init__(self, wrapped_model, threshold, target_class=1):
+        super().__init__(wrapped_model.domain, wrapped_model.original_domain)
+        self.name = f"{wrapped_model.name}, thresh={threshold:.2f}"
+        self.wrapped_model = wrapped_model
+        self.threshold = threshold
+        self.target_class = target_class
+
+    def __call__(self, data, ret=Model.Value):
+        probs = self.wrapped_model(data, ret=Model.Probs)
+        if ret == Model.Probs:
+            return probs
+        vals = probs[:, self.target_class].flatten() > self.threshold
+        if ret == Model.Value:
+            return vals
+        else:
+            return vals, probs