Add _n_inner_iter attribute for Logistic Regression (#1756)

* Add inner_iteraions_count option

* Add condition for onedal version

* Change required onedal version to 2024.3

* Remove _n_iter_iter from save attributes

* Add comment and test for logistic regression attributes presence

Author: avolkov-intel

onedal/linear_model/logistic_regression.cpp

@@ -95,8 +95,14 @@ auto get_onedal_result_options(const py::dict& params) {
             else if (match.str() == "iterations_count") {
                 onedal_options = onedal_options | result_options::iterations_count;
             }
-            else
+#if ONEDAL_VERSION >= 20240300
+            else if (match.str() == "inner_iterations_count") {
+                onedal_options = onedal_options | result_options::inner_iterations_count;
+            }
+#endif 
+            else {
                 ONEDAL_PARAM_DISPATCH_THROW_INVALID_VALUE(result_option);
+            }
         }
     }
     catch (std::regex_error&) {
@@ -207,6 +213,9 @@ void init_train_result(py::module_& m) {
                    .DEF_ONEDAL_PY_PROPERTY(intercept, result_t)
                    .DEF_ONEDAL_PY_PROPERTY(coefficients, result_t)
                    .DEF_ONEDAL_PY_PROPERTY(iterations_count, result_t)
+#if ONEDAL_VERSION >= 20240300
+                   .DEF_ONEDAL_PY_PROPERTY(inner_iterations_count, result_t)
+#endif
                    .DEF_ONEDAL_PY_PROPERTY(packed_coefficients, result_t)
                    .DEF_ONEDAL_PY_PROPERTY(result_options, result_t);
 }

onedal/linear_model/logistic_regression.py

@@ -19,7 +19,7 @@
 
 import numpy as np
 
-from daal4py.sklearn._utils import get_dtype, make2d
+from daal4py.sklearn._utils import daal_check_version, get_dtype, make2d
 
 from ..common._base import BaseEstimator as onedal_BaseEstimator
 from ..common._estimator_checks import _check_is_fitted
@@ -54,7 +54,11 @@ def _get_onedal_params(self, dtype=np.float32):
             "max_iter": self.max_iter,
             "C": self.C,
             "optimizer": self.solver,
-            "result_option": (intercept + "coefficients|iterations_count"),
+            "result_option": (
+                intercept
+                + "coefficients|iterations_count"
+                + ("|inner_iterations_count" if self.solver == "newton-cg" else "")
+            ),
         }
 
     def _fit(self, X, y, module, queue):
@@ -85,6 +89,10 @@ def _fit(self, X, y, module, queue):
         self._onedal_model = result.model
         self.n_iter_ = np.array([result.iterations_count])
 
+        # _n_inner_iter is the total number of cg-solver iterations
+        if daal_check_version((2024, "P", 300)) and self.solver == "newton-cg":
+            self._n_inner_iter = result.inner_iterations_count
+
         coeff = from_table(result.model.packed_coefficients)
         self.coef_, self.intercept_ = coeff[:, 1:], coeff[:, 0]
 

onedal/linear_model/tests/test_logistic_regression.py

@@ -40,6 +40,12 @@ def test_breast_cancer(queue, dtype):
         y_pred = model.predict(X_test, queue=queue)
         assert accuracy_score(y_test, y_pred) > 0.95
 
+        assert hasattr(model, "n_iter_")
+        assert hasattr(model, "coef_")
+        assert hasattr(model, "intercept_")
+        if daal_check_version((2024, "P", 300)):
+            assert hasattr(model, "_n_inner_iter")
+
     @pytest.mark.parametrize("queue", get_queues("gpu"))
     @pytest.mark.parametrize("dtype", [np.float32, np.float64])
     def test_pickle(queue, dtype):