Update progress bars in PGDs

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

Author: Beat Buesser

art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_numpy.py

@@ -30,6 +30,7 @@
 
 import numpy as np
 from scipy.stats import truncnorm
+from tqdm import trange
 
 from art.attacks.evasion.fast_gradient import FastGradientMethod
 from art.config import ART_NUMPY_DTYPE
@@ -266,10 +267,10 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
             adv_x_best = None
             rate_best = None
 
-            for _ in range(max(1, self.num_random_init)):
+            for _ in trange(max(1, self.num_random_init), desc="PGD - Random Initializations"):
                 adv_x = x.astype(ART_NUMPY_DTYPE)
 
-                for i_max_iter in range(self.max_iter):
+                for i_max_iter in trange(self.max_iter, desc="PGD - Iterations", leave=False):
                     adv_x = self._compute(
                         adv_x,
                         x,
@@ -313,7 +314,7 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
             # Start to compute adversarial examples
             adv_x = x.astype(ART_NUMPY_DTYPE)
 
-            for i_max_iter in range(self.max_iter):
+            for i_max_iter in trange(self.max_iter, desc="PGD - Iterations"):
                 adv_x = self._compute(
                     adv_x,
                     x,

art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py

@@ -29,6 +29,7 @@
 from typing import Optional, Union, TYPE_CHECKING
 
 import numpy as np
+from tqdm import trange, tqdm
 
 from art.config import ART_NUMPY_DTYPE
 from art.attacks.evasion.projected_gradient_descent.projected_gradient_descent_numpy import (
@@ -163,11 +164,11 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
         adv_x_best = None
         rate_best = None
 
-        for _ in range(max(1, self.num_random_init)):
+        for _ in trange(max(1, self.num_random_init), desc="PGD - Random Initializations"):
             adv_x = x.astype(ART_NUMPY_DTYPE)
 
             # Compute perturbation with batching
-            for (batch_id, batch_all) in enumerate(data_loader):
+            for (batch_id, batch_all) in enumerate(tqdm(data_loader, desc="PGD - Iterations", leave=False)):
                 if mask is not None:
                     (batch, batch_labels, mask_batch) = batch_all[0], batch_all[1], batch_all[2]
                 else:

art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_tensorflow_v2.py

@@ -29,6 +29,7 @@
 from typing import Optional, Union, TYPE_CHECKING
 
 import numpy as np
+from tqdm import trange, tqdm
 
 from art.config import ART_NUMPY_DTYPE
 from art.attacks.evasion.projected_gradient_descent.projected_gradient_descent_numpy import (
@@ -158,12 +159,12 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
         adv_x_best = None
         rate_best = None
 
-        for _ in range(max(1, self.num_random_init)):
+        for _ in trange(max(1, self.num_random_init), desc="PGD - Random Initializations"):
             adv_x = x.astype(ART_NUMPY_DTYPE)
             data_loader = iter(dataset)
 
             # Compute perturbation with batching
-            for (batch_id, batch_all) in enumerate(data_loader):
+            for (batch_id, batch_all) in enumerate(tqdm(data_loader, desc="PGD - Iterations", leave=False)):
                 if mask is not None:
                     (batch, batch_labels, mask_batch) = batch_all[0], batch_all[1], batch_all[2]
                 else:

art/defences/trainer/adversarial_trainer.py

@@ -120,7 +120,7 @@ def fit_generator(self, generator: "DataGenerator", nb_epochs: int = 20, **kwarg
         # Precompute adversarial samples for transferred attacks
         logged = False
         self._precomputed_adv_samples = []
-        for attack in tqdm(self.attacks, desc="Precompute adv samples"):
+        for attack in tqdm(self.attacks, desc="Precompute adversarial examples."):
             if "targeted" in attack.attack_params and attack.targeted:  # type: ignore
                 raise NotImplementedError("Adversarial training with targeted attacks is currently not implemented")
 

art/defences/trainer/adversarial_trainer_fbf_pytorch.py

@@ -27,6 +27,7 @@
 from typing import Optional, Tuple, Union, TYPE_CHECKING
 
 import numpy as np
+from tqdm import trange
 
 from art.config import ART_NUMPY_DTYPE
 from art.defences.trainer.adversarial_trainer_fbf import AdversarialTrainerFBF
@@ -92,9 +93,9 @@ def fit(
         def lr_schedule(t):
             return np.interp([t], [0, nb_epochs * 2 // 5, nb_epochs], [0, 0.21, 0])[0]
 
-        logger.info("Adversarial training FBF")
+        logger.info("Adversarial Training FBF")
 
-        for i_epoch in range(nb_epochs):
+        for i_epoch in trange(nb_epochs, desc="Adversarial Training FBF - Epochs"):
             # Shuffle the examples
             np.random.shuffle(ind)
             start_time = time.time()
@@ -160,9 +161,9 @@ def fit_generator(self, generator: "DataGenerator", nb_epochs: int = 20, **kwarg
         def lr_schedule(t):
             return np.interp([t], [0, nb_epochs * 2 // 5, nb_epochs], [0, 0.21, 0])[0]
 
-        logger.info("Adversarial training FBF")
+        logger.info("Adversarial Training FBF")
 
-        for i_epoch in range(nb_epochs):
+        for i_epoch in trange(nb_epochs, desc="Adversarial Training FBF - Epochs"):
             start_time = time.time()
             train_loss = 0.0
             train_acc = 0.0