Merge pull request #6921 from VesnaT/summary_of_fit

[ENH] Parameter Fitter: Basic implementation

Author: lanzagar

Orange/base.py

@@ -3,7 +3,7 @@
 from collections.abc import Iterable
 import re
 import warnings
-from typing import Callable, Dict, Optional
+from typing import Callable, Optional, NamedTuple, Type
 
 import numpy as np
 import scipy
@@ -88,6 +88,13 @@ class Learner(ReprableWithPreprocessors):
     #: fitting the model
     preprocessors = ()
 
+    class FittedParameter(NamedTuple):
+        name: str
+        label: str
+        type: Type
+        min: Optional[int] = None
+        max: Optional[int] = None
+
     # Note: Do not use this class attribute.
     #       It remains here for compatibility reasons.
     learner_adequacy_err_msg = ''
@@ -179,6 +186,10 @@ def active_preprocessors(self):
                 self.preprocessors is not type(self).preprocessors):
             yield from type(self).preprocessors
 
+    @property
+    def fitted_parameters(self) -> list:
+        return []
+
     # pylint: disable=no-self-use
     def incompatibility_reason(self, _: Domain) -> Optional[str]:
         """Return None if a learner can fit domain or string explaining why it can not."""
@@ -883,5 +894,5 @@ def __init__(self, preprocessors=None, **kwargs):
         self.params = kwargs
 
     @SklLearner.params.setter
-    def params(self, values: Dict):
+    def params(self, values: dict):
         self._params = values

Orange/evaluation/testing.py

@@ -25,7 +25,7 @@ def _identity(x):
 
 
 def _mp_worker(fold_i, train_data, test_data, learner_i, learner,
-               store_models):
+               store_models, suppresses_exceptions=True):
     predicted, probs, model, failed = None, None, None, False
     train_time, test_time = None, None
     try:
@@ -45,6 +45,8 @@ def _mp_worker(fold_i, train_data, test_data, learner_i, learner,
         test_time = time() - t0
     # Different models can fail at any time raising any exception
     except Exception as ex:  # pylint: disable=broad-except
+        if not suppresses_exceptions:
+            raise ex
         failed = ex
     return _MpResults(fold_i, learner_i, store_models and model,
                       failed, len(test_data), predicted, probs,
@@ -96,6 +98,7 @@ def __init__(self, data=None, *,
                  row_indices=None, folds=None, score_by_folds=True,
                  learners=None, models=None, failed=None,
                  actual=None, predicted=None, probabilities=None,
+                 # pylint: disable=unused-argument
                  store_data=None, store_models=None,
                  train_time=None, test_time=None):
         """
@@ -426,7 +429,8 @@ def fit(self, *args, **kwargs):
              DeprecationWarning)
         return self(*args, **kwargs)
 
-    def __call__(self, data, learners, preprocessor=None, *, callback=None):
+    def __call__(self, data, learners, preprocessor=None, *, callback=None,
+                 suppresses_exceptions=True):
         """
         Args:
             data (Orange.data.Table): data to be used (usually split) into
@@ -435,6 +439,7 @@ def __call__(self, data, learners, preprocessor=None, *, callback=None):
             preprocessor (Orange.preprocess.Preprocess): preprocessor applied
                 on training data
             callback (Callable): a function called to notify about the progress
+            suppresses_exceptions (bool): suppress the exceptions if True
 
         Returns:
             results (Result): results of testing
@@ -457,7 +462,10 @@ def __call__(self, data, learners, preprocessor=None, *, callback=None):
         part_results = []
         parts = np.linspace(.0, .99, len(learners) * len(indices) + 1)[1:]
         for progress, part in zip(parts, args_iter):
-            part_results.append(_mp_worker(*(part + ())))
+            part_results.append(
+                _mp_worker(*(part + ()),
+                           suppresses_exceptions=suppresses_exceptions)
+            )
             callback(progress)
         callback(1)
 
@@ -723,7 +731,7 @@ def __new__(cls, data=None, test_data=None, learners=None,
             test_data=test_data, **kwargs)
 
     def __call__(self, data, test_data, learners, preprocessor=None,
-                 *, callback=None):
+                 *, callback=None, suppresses_exceptions=True):
         """
         Args:
             data (Orange.data.Table): training data
@@ -732,6 +740,7 @@ def __call__(self, data, test_data, learners, preprocessor=None,
             preprocessor (Orange.preprocess.Preprocess): preprocessor applied
                 on training data
             callback (Callable): a function called to notify about the progress
+            suppresses_exceptions (bool): suppress the exceptions if True
 
         Returns:
             results (Result): results of testing
@@ -746,7 +755,7 @@ def __call__(self, data, test_data, learners, preprocessor=None,
         for (learner_i, learner) in enumerate(learners):
             part_results.append(
                 _mp_worker(0, train_data, test_data, learner_i, learner,
-                           self.store_models))
+                           self.store_models, suppresses_exceptions))
             callback((learner_i + 1) / len(learners))
         callback(1)
 
@@ -778,13 +787,14 @@ def __new__(cls, data=None, learners=None, preprocessor=None, **kwargs):
             **kwargs)
 
     def __call__(self, data, learners, preprocessor=None, *, callback=None,
-                 **kwargs):
+                 suppresses_exceptions=True, **kwargs):
         kwargs.setdefault("test_data", data)
         # if kwargs contains anything besides test_data, this will be detected
         # (and complained about) by super().__call__
         return super().__call__(
             data=data, learners=learners, preprocessor=preprocessor,
-            callback=callback, **kwargs)
+            callback=callback, suppresses_exceptions=suppresses_exceptions,
+            **kwargs)
 
 
 def sample(table, n=0.7, stratified=False, replace=False,

Orange/modelling/randomforest.py

@@ -1,4 +1,4 @@
-from Orange.base import RandomForestModel
+from Orange.base import RandomForestModel, Learner
 from Orange.classification import RandomForestLearner as RFClassification
 from Orange.data import Variable
 from Orange.modelling import SklFitter
@@ -24,3 +24,8 @@ class RandomForestLearner(SklFitter, _FeatureScorerMixin):
                 'regression': RFRegression}
 
     __returns__ = RandomForestModel
+
+    @property
+    def fitted_parameters(self) -> list[Learner.FittedParameter]:
+        return [self.FittedParameter("n_estimators", "Number of trees",
+                                     int, 1, None)]

Orange/regression/pls.py

@@ -1,10 +1,9 @@
-from typing import Tuple
-
 import numpy as np
 import scipy.stats as ss
 import sklearn.cross_decomposition as skl_pls
 from sklearn.preprocessing import StandardScaler
 
+from Orange.base import Learner
 from Orange.data import Table, Domain, Variable, \
     ContinuousVariable, StringVariable
 from Orange.data.util import get_unique_names, SharedComputeValue
@@ -163,11 +162,11 @@ def coefficients_table(self):
         return coef_table
 
     @property
-    def rotations(self) -> Tuple[np.ndarray, np.ndarray]:
+    def rotations(self) -> tuple[np.ndarray, np.ndarray]:
         return self.skl_model.x_rotations_, self.skl_model.y_rotations_
 
     @property
-    def loadings(self) -> Tuple[np.ndarray, np.ndarray]:
+    def loadings(self) -> tuple[np.ndarray, np.ndarray]:
         return self.skl_model.x_loadings_, self.skl_model.y_loadings_
 
     def residuals_normal_probability(self, data: Table) -> Table:
@@ -256,6 +255,11 @@ def incompatibility_reason(self, domain):
                     reason = "Only numeric target variables expected."
         return reason
 
+    @property
+    def fitted_parameters(self) -> list[Learner.FittedParameter]:
+        return [self.FittedParameter("n_components", "Components",
+                                     int, 1, None)]
+
 
 if __name__ == '__main__':
     import Orange

Orange/regression/tests/test_pls.py

@@ -21,6 +21,11 @@ def table(rows, attr, variables):
 
 
 class TestPLSRegressionLearner(unittest.TestCase):
+    def test_fitted_parameters(self):
+        fitted_parameters = PLSRegressionLearner().fitted_parameters
+        self.assertIsInstance(fitted_parameters, list)
+        self.assertEqual(len(fitted_parameters), 1)
+
     def test_allow_y_dim(self):
         """ The current PLS version allows only a single Y dimension. """
         learner = PLSRegressionLearner(n_components=2)