Optimise test time

Signed-off-by: Beat Buesser <[email protected]>

Author: beat-buesser

tests/defences/trainer/test_adversarial_trainer_awp_pytorch.py

@@ -74,9 +74,9 @@ def _get_adv_trainer_awptrades():
                 norm=np.inf,
                 eps=0.2,
                 eps_step=0.02,
-                max_iter=20,
+                max_iter=5,
                 targeted=False,
-                num_random_init=1,
+                num_random_init=0,
                 batch_size=128,
                 verbose=False,
             )
@@ -141,7 +141,7 @@ def test_adversarial_trainer_awppgd_pytorch_fit_and_predict(get_adv_trainer_awpp
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=5, validation_data=(x_train_mnist, y_train_mnist))
 
 
 @pytest.mark.only_with_platform("pytorch")
@@ -171,7 +171,7 @@ def test_adversarial_trainer_awptrades_pytorch_fit_and_predict(
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20)
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=5)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -188,7 +188,7 @@ def test_adversarial_trainer_awptrades_pytorch_fit_and_predict(
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=5, validation_data=(x_train_mnist, y_train_mnist))
 
 
 @pytest.mark.only_with_platform("pytorch")
@@ -219,7 +219,7 @@ def test_adversarial_trainer_awppgd_pytorch_fit_generator_and_predict(
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit_generator(generator=generator, nb_epochs=20)
+    trainer.fit_generator(generator=generator, nb_epochs=5)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -236,7 +236,7 @@ def test_adversarial_trainer_awppgd_pytorch_fit_generator_and_predict(
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit_generator(generator=generator, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit_generator(generator=generator, nb_epochs=2, validation_data=(x_train_mnist, y_train_mnist))
 
 
 @pytest.mark.only_with_platform("pytorch")
@@ -267,7 +267,7 @@ def test_adversarial_trainer_awptrades_pytorch_fit_generator_and_predict(
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit_generator(generator=generator, nb_epochs=20)
+    trainer.fit_generator(generator=generator, nb_epochs=5)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -284,4 +284,4 @@ def test_adversarial_trainer_awptrades_pytorch_fit_generator_and_predict(
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit_generator(generator=generator, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit_generator(generator=generator, nb_epochs=2, validation_data=(x_train_mnist, y_train_mnist))

tests/defences/trainer/test_adversarial_trainer_oaat_pytorch.py

@@ -59,7 +59,7 @@ def _get_adv_trainer_oaat():
                 "alternate_iter_eps": 0.15,
                 "swa_save_epoch": 0,
                 "list_swa_epoch": [0, 15, 15],
-                "max_iter": 20,
+                "max_iter": 5,
                 "models_path": None,
                 "load_swa_model_tau": 0.995,
                 "layer_names_activation": ["conv"],
@@ -120,7 +120,7 @@ def test_adversarial_trainer_oaat_pytorch_fit_and_predict(get_adv_trainer_oaat,
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20, batch_size=16)
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=5, batch_size=16)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -138,7 +138,7 @@ def test_adversarial_trainer_oaat_pytorch_fit_and_predict(get_adv_trainer_oaat,
     assert accuracy_new > 0.32
 
     trainer.fit(
-        x_train_mnist, y_train_mnist, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist), batch_size=16
+        x_train_mnist, y_train_mnist, nb_epochs=2, validation_data=(x_train_mnist, y_train_mnist), batch_size=16
     )
 
 
@@ -170,7 +170,7 @@ def test_adversarial_trainer_oaat_pytorch_fit_generator_and_predict(
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit_generator(generator=generator, nb_epochs=20)
+    trainer.fit_generator(generator=generator, nb_epochs=10)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -187,4 +187,4 @@ def test_adversarial_trainer_oaat_pytorch_fit_generator_and_predict(
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit_generator(generator=generator, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit_generator(generator=generator, nb_epochs=2, validation_data=(x_train_mnist, y_train_mnist))

tests/defences/trainer/test_adversarial_trainer_trades_pytorch.py

@@ -40,7 +40,7 @@ def _get_adv_trainer():
                 norm=np.inf,
                 eps=0.3,
                 eps_step=0.03,
-                max_iter=20,
+                max_iter=5,
                 targeted=False,
                 num_random_init=1,
                 batch_size=128,
@@ -89,7 +89,7 @@ def test_adversarial_trainer_trades_pytorch_fit_and_predict(get_adv_trainer, fix
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20)
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=5)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":
@@ -106,7 +106,7 @@ def test_adversarial_trainer_trades_pytorch_fit_and_predict(get_adv_trainer, fix
     assert accuracy == 0.32
     assert accuracy_new > 0.32
 
-    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=20, validation_data=(x_train_mnist, y_train_mnist))
+    trainer.fit(x_train_mnist, y_train_mnist, nb_epochs=2, validation_data=(x_train_mnist, y_train_mnist))
 
 
 @pytest.mark.only_with_platform("pytorch")
@@ -136,7 +136,7 @@ def test_adversarial_trainer_trades_pytorch_fit_generator_and_predict(
     else:
         accuracy = np.sum(predictions == y_test_mnist) / x_test_mnist.shape[0]
 
-    trainer.fit_generator(generator=generator, nb_epochs=20)
+    trainer.fit_generator(generator=generator, nb_epochs=5)
     predictions_new = np.argmax(trainer.predict(x_test_mnist), axis=1)
 
     if label_format == "one_hot":