Collect features 3 times (after-load, after-lp, after-mip)

miplearn/instance/base.py

@@ -33,6 +33,9 @@ class Instance(ABC, EnforceOverrides):
     def __init__(self) -> None:
         self.training_data: List[TrainingSample] = []
         self.features: Features = Features()
+        self.features_after_load: List[Features] = []
+        self.features_after_lp: List[Features] = []
+        self.features_after_mip: List[Features] = []
 
     @abstractmethod
     def to_model(self) -> Any:

miplearn/instance/picklegz.py

@@ -123,6 +123,9 @@ class PickleGzInstance(Instance):
             self.instance = obj
             self.features = self.instance.features
             self.training_data = self.instance.training_data
+            self.features_after_load = self.instance.features_after_load
+            self.features_after_lp = self.instance.features_after_lp
+            self.features_after_mip = self.instance.features_after_mip
 
     @overrides
     def free(self) -> None:

miplearn/solvers/learning.py

@@ -153,11 +153,12 @@ class LearningSolver:
         assert isinstance(self.internal_solver, InternalSolver)
         self.internal_solver.set_instance(instance, model)
 
-        # Extract features
+        # Extract features (after-load)
         # -------------------------------------------------------
-        if instance.features.instance is None:
-            logger.info("Extracting features...")
-            FeaturesExtractor(self.internal_solver).extract(instance)
+        logger.info("Extracting features (after-load)...")
+        features = FeaturesExtractor(self.internal_solver).extract(instance)
+        instance.features.__dict__ = features.__dict__
+        instance.features_after_load.append(features)
 
         callback_args = (
             self,
@@ -186,6 +187,12 @@ class LearningSolver:
             for component in self.components.values():
                 component.after_solve_lp(*callback_args)
 
+            # Extract features (after-lp)
+            # -------------------------------------------------------
+            logger.info("Extracting features (after-lp)...")
+            features = FeaturesExtractor(self.internal_solver).extract(instance)
+            instance.features_after_lp.append(features)
+
         # Callback wrappers
         # -------------------------------------------------------
         def iteration_cb_wrapper() -> bool:
@@ -242,6 +249,12 @@ class LearningSolver:
         )
         stats["Mode"] = self.mode
 
+        # Extract features (after-mip)
+        # -------------------------------------------------------
+        logger.info("Extracting features (after-mip)...")
+        features = FeaturesExtractor(self.internal_solver).extract(instance)
+        instance.features_after_mip.append(features)
+
         # Add some information to training_sample
         # -------------------------------------------------------
         training_sample.lower_bound = stats["Lower bound"]