[FIX] Fix missing base estimators when calling load (#45)

* fix save and load

* remove randomness in unit tests

CHANGELOG.rst

@@ -4,6 +4,7 @@ Changelog
 [Ver 0.1.*]
 -----------
 
+* |Fix| Fix missing base estimators when calling :meth:`load()` for all ensembles | `@xuyxu <https://github.com/xuyxu>`__
 * |MajorFeature| Add methods on model deserialization :meth:`load()` for all ensembles | `@mttgdd <https://github.com/mttgdd>`__
 
 [Beta]

torchensemble/tests/test_all_models.py

@@ -5,6 +5,7 @@
 from torch.utils.data import TensorDataset, DataLoader
 
 import torchensemble
+from torchensemble.utils import io
 from torchensemble.utils.logging import set_logger
 
 
@@ -24,6 +25,10 @@
            torchensemble.AdversarialTrainingRegressor]
 
 
+# Remove randomness
+np.random.seed(0)
+torch.manual_seed(0)
+
 set_logger("pytest_all_models")
 
 
@@ -66,12 +71,10 @@ def forward(self, X):
 
 # Testing data
 X_test = torch.Tensor(np.array(([0.5, 0.5],
-                                [0.6, 0.6],
-                                [0.7, 0.7],
-                                [0.8, 0.8])))
+                                [0.6, 0.6])))
 
-y_test_clf = torch.LongTensor(np.array(([1, 1, 0, 0])))
-y_test_reg = torch.FloatTensor(np.array(([0.5, 0.6, 0.7, 0.8])))
+y_test_clf = torch.LongTensor(np.array(([1, 0])))
+y_test_reg = torch.FloatTensor(np.array(([0.5, 0.6])))
 y_test_reg = y_test_reg.view(-1, 1)
 
 
@@ -94,9 +97,9 @@ def test_clf(clf):
 
     # Prepare data
     train = TensorDataset(X_train, y_train_clf)
-    train_loader = DataLoader(train, batch_size=2)
+    train_loader = DataLoader(train, batch_size=2, shuffle=False)
     test = TensorDataset(X_test, y_test_clf)
-    test_loader = DataLoader(test, batch_size=2)
+    test_loader = DataLoader(test, batch_size=2, shuffle=False)
 
     # Snapshot ensemble needs more epochs
     if isinstance(model, torchensemble.SnapshotEnsembleClassifier):
@@ -109,7 +112,15 @@ def test_clf(clf):
               save_model=True)
 
     # Test
-    model.predict(test_loader)
+    prev_acc = model.predict(test_loader)
+
+    # Reload
+    new_model = clf(estimator=MLP_clf, n_estimators=n_estimators, cuda=False)
+    io.load(new_model)
+
+    post_acc = new_model.predict(test_loader)
+
+    assert prev_acc == post_acc  # ensure the same performance
 
 
 @pytest.mark.parametrize("reg", all_reg)
@@ -131,9 +142,9 @@ def test_reg(reg):
 
     # Prepare data
     train = TensorDataset(X_train, y_train_reg)
-    train_loader = DataLoader(train, batch_size=2)
+    train_loader = DataLoader(train, batch_size=2, shuffle=False)
     test = TensorDataset(X_test, y_test_reg)
-    test_loader = DataLoader(test, batch_size=2)
+    test_loader = DataLoader(test, batch_size=2, shuffle=False)
 
     # Snapshot ensemble needs more epochs
     if isinstance(model, torchensemble.SnapshotEnsembleRegressor):
@@ -146,7 +157,15 @@ def test_reg(reg):
               save_model=True)
 
     # Test
-    model.predict(test_loader)
+    prev_mse = model.predict(test_loader)
+
+    # Reload
+    new_model = reg(estimator=MLP_reg, n_estimators=n_estimators, cuda=False)
+    io.load(new_model)
+
+    post_mse = new_model.predict(test_loader)
+
+    assert prev_mse == post_mse  # ensure the same performance
 
 
 @pytest.mark.parametrize("method", all_clf + all_reg)

torchensemble/utils/io.py

@@ -14,7 +14,11 @@ def save(model, save_dir, logger):
     filename = "{}_{}_{}_ckpt.pth".format(type(model).__name__,
                                           model.base_estimator_.__name__,
                                           model.n_estimators)
-    state = {"model": model.state_dict()}
+
+    # The real number of base estimators in some ensembles is not same as
+    # `n_estimators`.
+    state = {"n_estimators": len(model.estimators_),
+             "model": model.state_dict()}
     save_dir = os.path.join(save_dir, filename)
 
     logger.info("Saving the model to `{}`".format(save_dir))
@@ -39,4 +43,11 @@ def load(model, save_dir="./", logger=None):
     if logger:
         logger.info("Loading the model from `{}`".format(save_dir))
 
-    model.load_state_dict(torch.load(save_dir)["model"])
+    state = torch.load(save_dir)
+    n_estimators = state["n_estimators"]
+    model_params = state["model"]
+
+    # Pre-allocate and load all base estimators
+    for _ in range(n_estimators):
+        model.estimators_.append(model._make_estimator())
+    model.load_state_dict(model_params)