Fix DatabaseScaler plugin

Author: Davide-Miotti

ezyrb/plugin/scaler.py

@@ -1,4 +1,4 @@
-""" Module for Scaler plugin """
+"""Module for Scaler plugin"""
 
 from .plugin import Plugin
 
@@ -19,9 +19,9 @@ class DatabaseScaler(Plugin):
         applied at the full order ('full') or at the reduced one ('reduced').
     :param {'parameters', 'snapshots'} params: define if the rescaling has to
         be applied to the parameters or to the snapshots.
-    
+
     :Example:
-    
+
         >>> from ezyrb import ReducedOrderModel as ROM
         >>> from ezyrb import POD, RBF, Database
         >>> from ezyrb.plugin import DatabaseScaler
@@ -33,10 +33,11 @@ class DatabaseScaler(Plugin):
         >>> rom = ROM(db, pod, rbf, plugins=[scaler])
         >>> rom.fit()
     """
+
     def __init__(self, scaler, mode, target) -> None:
         """
         Initialize the DatabaseScaler plugin.
-        
+
         :param scaler: Scaler object with fit, transform, and inverse_transform methods.
         :param str mode: 'full' or 'reduced' - where to apply the scaling.
         :param str target: 'parameters' or 'snapshots' - what to scale.
@@ -46,7 +47,7 @@ def __init__(self, scaler, mode, target) -> None:
         self.scaler = scaler
         self.mode = mode
         self.target = target
-        
+
     @property
     def target(self):
         """
@@ -58,7 +59,7 @@ def target(self):
 
     @target.setter
     def target(self, new_target):
-        if new_target not in ['snapshots', 'parameters']:
+        if new_target not in ["snapshots", "parameters"]:
             raise ValueError
 
         self._target = new_target
@@ -74,102 +75,152 @@ def mode(self):
 
     @mode.setter
     def mode(self, new_mode):
-        if new_mode not in ['full', 'reduced']:
+        if new_mode not in ["full", "reduced"]:
             raise ValueError
 
         self._mode = new_mode
 
     def _select_matrix(self, db):
         """
         Helper function to select the proper matrix to rescale.
-        
+
         :param Database db: The database object.
         :return: The selected matrix (parameters or snapshots).
         """
-        return getattr(db, f'{self.target}_matrix')
+        return getattr(db, f"{self.target}_matrix")
 
-    def rom_preprocessing(self, rom):
+    # =========================================================================
+    # MODE = 'FULL' - Scaling applied at full order (before reduction or after prediction)
+    # =========================================================================
+
+    def fit_before_reduction(self, rom):
         """
-        Apply scaling to the reduced database before ROM processing.
-        
+        Apply scaling before POD reduction when mode='full'.
+        Scales the full-order database before reduction.
+
         :param ReducedOrderModel rom: The ROM instance.
         """
-        if self.mode != 'reduced':
+        if self.mode != "full":
             return
 
-        db = rom._reduced_database
+        db = rom.train_full_database
 
         self.scaler.fit(self._select_matrix(db))
 
-        if self.target == 'parameters':
+        if self.target == "parameters":
             new_db = type(db)(
                 self.scaler.transform(self._select_matrix(db)),
-                db.snapshots_matrix
+                db.snapshots_matrix,
             )
         else:
             new_db = type(db)(
                 db.parameters_matrix,
                 self.scaler.transform(self._select_matrix(db)),
             )
 
-        rom._reduced_database = new_db
+        rom.train_full_database = new_db
 
-    def fom_preprocessing(self, rom):
-        if self.mode != 'full':
-            return
+    def predict_postprocessing(self, rom):
+        """
+        Inverse transform scaled data after prediction when mode='full'.
+        Restores original scale to the full-order predicted database.
 
-        db = rom._full_database
+        :param ReducedOrderModel rom: The ROM instance.
+        """
+        if self.mode != "full":
+            return
 
-        self.scaler.fit(self._select_matrix(db))
+        db = rom.predicted_full_database
 
-        if self.target == 'parameters':
+        if self.target == "parameters":
             new_db = type(db)(
-                self.scaler.transform(self._select_matrix(db)),
-                db.snapshots_matrix
+                self.scaler.inverse_transform(self._select_matrix(db)),
+                db.snapshots_matrix,
             )
         else:
             new_db = type(db)(
                 db.parameters_matrix,
-                self.scaler.transform(self._select_matrix(db)),
+                self.scaler.inverse_transform(self._select_matrix(db)),
             )
 
-        rom._full_database = new_db
+        rom.predicted_full_database = new_db
+
+    # =========================================================================
+    # MODE = 'REDUCED' - Scaling applied at reduced order (before/after approximation)
+    # =========================================================================
 
-    def fom_postprocessing(self, rom):
+    def fit_before_approximation(self, rom):
+        """
+        Apply scaling before approximation training when mode='reduced'.
+        Scales the reduced database before approximation training.
 
-        if self.mode != 'full':
+        :param ReducedOrderModel rom: The ROM instance.
+        """
+        if self.mode != "reduced":
             return
 
-        db = rom._full_database
+        db = rom.train_reduced_database
 
-        if self.target == 'parameters':
+        self.scaler.fit(self._select_matrix(db))
+
+        if self.target == "parameters":
             new_db = type(db)(
-                self.scaler.inverse_transform(self._select_matrix(db)),
-                db.snapshots_matrix
+                self.scaler.transform(self._select_matrix(db)),
+                db.snapshots_matrix,
             )
         else:
             new_db = type(db)(
                 db.parameters_matrix,
-                self.scaler.inverse_transform(self._select_matrix(db)),
+                self.scaler.transform(self._select_matrix(db)),
             )
 
-        rom._full_database = new_db
+        rom.train_reduced_database = new_db
 
-    def rom_postprocessing(self, rom):
-        if self.mode != 'reduced':
+    def predict_after_approximation(self, rom):
+        """
+        Inverse transform scaled data after approximation when mode='reduced'.
+        Restores original scale to the reduced predicted database.
+
+        :param ReducedOrderModel rom: The ROM instance.
+        """
+        if self.mode != "reduced":
             return
 
-        db = rom._reduced_database
+        db = rom.predict_reduced_database
 
-        if self.target == 'parameters':
+        if self.target == "parameters":
             new_db = type(db)(
                 self.scaler.inverse_transform(self._select_matrix(db)),
-                db.snapshots_matrix
+                db.snapshots_matrix,
             )
         else:
             new_db = type(db)(
                 db.parameters_matrix,
                 self.scaler.inverse_transform(self._select_matrix(db)),
             )
-            
-        rom._reduced_database = new_db
+
+        rom.predict_reduced_database = new_db
+
+    # =========================================================================
+    # PREDICT - Scaling input parameters before approximation (both modes)
+    # =========================================================================
+
+    def predict_before_approximation(self, rom):
+        """
+        Transform (scale) input parameters before approximation if target='parameters'.
+        This ensures parameters are scaled to match the training data.
+        Applied during prediction for both 'full' and 'reduced' modes.
+
+        :param ReducedOrderModel rom: The ROM instance.
+        """
+        if self.target != "parameters":
+            return
+
+        db = rom.predict_reduced_database
+        transformed_params = self.scaler.transform(self._select_matrix(db))
+
+        # During prediction, snapshots are None (not yet predicted)
+        # Database constructor handles None snapshots: creates [None] * len(parameters)
+        new_db = type(db)(transformed_params, None)
+
+        rom.predict_reduced_database = new_db

tests/test_scaler.py

@@ -8,40 +8,124 @@
 
 from sklearn.preprocessing import StandardScaler, MinMaxScaler
 
-snapshots = np.load('tests/test_datasets/p_snapshots.npy').T
-pred_sol_tst = np.load('tests/test_datasets/p_predsol.npy').T
-pred_sol_gpr = np.load('tests/test_datasets/p_predsol_gpr.npy').T
-param = np.array([[-.5, -.5], [.5, -.5], [.5, .5], [-.5, .5]])
+snapshots = np.load("tests/test_datasets/p_snapshots.npy").T
+pred_sol_tst = np.load("tests/test_datasets/p_predsol.npy").T
+pred_sol_gpr = np.load("tests/test_datasets/p_predsol_gpr.npy").T
+param = np.array([[-0.5, -0.5], [0.5, -0.5], [0.5, 0.5], [-0.5, 0.5]])
 
 
 def test_constructor():
     pod = POD()
     import torch
+
     rbf = RBF()
-    #rbf = ANN([10, 10], function=torch.nn.Softplus(), stop_training=[1000])
+    # rbf = ANN([10, 10], function=torch.nn.Softplus(), stop_training=[1000])
     db = Database(param, snapshots.T)
     # rom = ROM(db, pod, rbf, plugins=[DatabaseScaler(StandardScaler(), 'full', 'snapshots')])
-    rom = ROM(db, pod, rbf, plugins=[
-        DatabaseScaler(StandardScaler(), 'reduced', 'parameters'),
-        DatabaseScaler(StandardScaler(), 'reduced', 'snapshots')
-    ])
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[
+            DatabaseScaler(StandardScaler(), "reduced", "parameters"),
+            DatabaseScaler(StandardScaler(), "reduced", "snapshots"),
+        ],
+    )
+    rom.fit()
+    assert rom is not None
+
+
+def test_scaler_reduced_snapshots():
+    """Test that StandardScaler on reduced snapshots produces mean=0 and std=1"""
+    pod = POD()
+    rbf = RBF()
+    db = Database(param, snapshots.T)
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[DatabaseScaler(StandardScaler(), "reduced", "snapshots")],
+    )
+    rom.fit()
+
+    # Check that the scaled reduced snapshots have mean ≈ 0 and std ≈ 1
+    scaled_snapshots = rom.train_reduced_database.snapshots_matrix
+    np.testing.assert_allclose(np.mean(scaled_snapshots, axis=0), 0, atol=1e-7)
+    np.testing.assert_allclose(np.std(scaled_snapshots, axis=0), 1, atol=1e-7)
+
+
+def test_scaler_reduced_parameters():
+    """Test that StandardScaler on reduced parameters produces mean=0 and std=1"""
+    pod = POD()
+    rbf = RBF()
+    db = Database(param, snapshots.T)
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[DatabaseScaler(StandardScaler(), "reduced", "parameters")],
+    )
+    rom.fit()
+
+    # Check that the scaled reduced parameters have mean ≈ 0 and std ≈ 1
+    scaled_params = rom.train_reduced_database.parameters_matrix
+    np.testing.assert_allclose(np.mean(scaled_params, axis=0), 0, atol=1e-7)
+    np.testing.assert_allclose(np.std(scaled_params, axis=0), 1, atol=1e-7)
+
+
+def test_scaler_full_snapshots():
+    """Test that StandardScaler on full snapshots produces mean=0 and std=1"""
+    pod = POD()
+    rbf = RBF()
+    db = Database(param, snapshots.T)
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[DatabaseScaler(StandardScaler(), "full", "snapshots")],
+    )
     rom.fit()
-   
-    
+
+    # Check that the scaled full snapshots have mean ≈ 0 and std ≈ 1
+    scaled_snapshots = rom.train_full_database.snapshots_matrix
+    np.testing.assert_allclose(np.mean(scaled_snapshots, axis=0), 0, atol=2e-6)
+    np.testing.assert_allclose(np.std(scaled_snapshots, axis=0), 1, atol=2e-6)
+
+
+def test_scaler_full_parameters():
+    """Test that StandardScaler on full parameters produces mean=0 and std=1"""
+    pod = POD()
+    rbf = RBF()
+    db = Database(param, snapshots.T)
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[DatabaseScaler(StandardScaler(), "full", "parameters")],
+    )
+    rom.fit()
+
+    # Check that the scaled full parameters have mean ≈ 0 and std ≈ 1
+    scaled_params = rom.train_full_database.parameters_matrix
+    np.testing.assert_allclose(np.mean(scaled_params, axis=0), 0, atol=2e-6)
+    np.testing.assert_allclose(np.std(scaled_params, axis=0), 1, atol=2e-6)
 
 
 def test_values():
     pod = POD()
     rbf = RBF()
     db = Database(param, snapshots.T)
-    rom = ROM(db, pod, rbf, plugins=[
-        DatabaseScaler(StandardScaler(), 'reduced', 'snapshots'),
-        DatabaseScaler(StandardScaler(), 'full', 'parameters')
-    ])
+    rom = ROM(
+        db,
+        pod,
+        rbf,
+        plugins=[
+            DatabaseScaler(StandardScaler(), "reduced", "snapshots"),
+            DatabaseScaler(StandardScaler(), "full", "parameters"),
+        ],
+    )
     rom.fit()
     test_param = param[2]
     truth_sol = db.snapshots_matrix[2]
     predicted_sol = rom.predict(test_param)[0]
-    np.testing.assert_allclose(predicted_sol, truth_sol, 
-            rtol=1e-5, atol=1e-5)
-
+    np.testing.assert_allclose(predicted_sol, truth_sol, rtol=1e-5, atol=1e-5)