fix: bug in doubling estimators_ if model loaded from save_dir + feature for having callback after each epoch (#166)

* fix: dont instantiate if estimators loaded from save_dir

* feat: add on_epoch_end_cb to call at end of each epoch

---------

Co-authored-by: [email protected] <[email protected]>

Author: h2soheili

torchensemble/soft_gradient_boosting.py

@@ -209,11 +209,14 @@ def fit(
         test_loader=None,
         save_model=True,
         save_dir=None,
+        on_epoch_end_cb=None
     ):
 
         # Instantiate base estimators and set attributes
-        for _ in range(self.n_estimators):
-            self.estimators_.append(self._make_estimator())
+        # dont instantiate if estimators loaded from save_dir
+        if len(self.estimators_) != self.n_estimators:
+            for _ in range(self.n_estimators):
+                self.estimators_.append(self._make_estimator())
         self._validate_parameters(epochs, log_interval)
         self.n_outputs = self._decide_n_outputs(train_loader)
 
@@ -295,6 +298,9 @@ def fit(
                 else:
                     scheduler.step()
 
+            # Call on epoch end
+            if on_epoch_end_cb:
+                on_epoch_end_cb(epoch)
         if save_model and not test_loader:
             io.save(self, save_dir, self.logger)
 
@@ -390,6 +396,7 @@ def fit(
         test_loader=None,
         save_model=True,
         save_dir=None,
+        on_epoch_end_cb=None
     ):
         super().fit(
             train_loader=train_loader,
@@ -399,6 +406,7 @@ def fit(
             test_loader=test_loader,
             save_model=save_model,
             save_dir=save_dir,
+            on_epoch_end_cb=on_epoch_end_cb,
         )
 
     @torchensemble_model_doc(