examples: update feature squeezing example to task engine

examples/task-plugins/dioptra_custom/feature_squeezing/cw_inf_plugin.py

@@ -59,7 +59,7 @@
 def create_adversarial_cw_inf_dataset(
     data_dir: str,
     model_name: str,
-    model_version: str,
+    model_version: int,
     learning_rate: float,
     max_iter: int,
     verbose: str,
@@ -74,7 +74,7 @@ def create_adversarial_cw_inf_dataset(
     confidence: float = 0.0,
     **kwargs,
 ):
-    model_name = model_name
+    model_name = model_name + "/" + str(model_version)
     LOGGER.info("Model Selected: ", model_name=model_name)
     color_mode: str = "color" if image_size[2] == 3 else "grayscale"
     target_size: Tuple[int, int] = image_size[:2]

examples/task-plugins/dioptra_custom/feature_squeezing/cw_l2_plugin.py

@@ -60,7 +60,7 @@
 def create_adversarial_cw_l2_dataset(
     data_dir: str,
     model_name: str,
-    model_version: str,
+    model_version: int,
     confidence: float,
     targeted: bool,
     learning_rate: float,
@@ -76,7 +76,7 @@ def create_adversarial_cw_l2_dataset(
     image_size: Tuple[int, int] = (28, 28),
     **kwargs,
 ):
-    model_name = model_name + "/" + model_version
+    model_name = model_name + "/" + str(model_version)
     LOGGER.info("Model Selected: ", model_name=model_name)
     color_mode: str = "color" if image_size[2] == 3 else "grayscale"
     target_size: Tuple[int, int] = image_size[:2]

examples/task-plugins/dioptra_custom/feature_squeezing/deepfool_plugin.py

@@ -73,7 +73,7 @@ def evaluate_classification_metrics(classifier, adv_ds):
 def create_adversarial_deepfool_dataset(
     data_dir: str,
     model_name: str,
-    model_version: str,
+    model_version: int,
     epsilon: float,
     nb_grads: int,
     max_iter: int,
@@ -86,7 +86,7 @@ def create_adversarial_deepfool_dataset(
     color_mode: str = "rgb",
     **kwargs,
 ):
-    model_name = model_name + "/" + model_version
+    model_name = model_name + "/" + str(model_version)
     attack = DeepFool(
         classifier=keras_classifier,
         batch_size=batch_size,

examples/task-plugins/dioptra_custom/feature_squeezing/jsma_plugin.py

@@ -59,7 +59,7 @@
 def create_adversarial_jsma_dataset(
     data_dir: str,
     model_name: str,
-    model_version: str,
+    model_version: int,
     theta: float,
     gamma: float,
     keras_classifier: KerasClassifier,
@@ -68,11 +68,11 @@ def create_adversarial_jsma_dataset(
     batch_size: int = 32,
     label_mode: str = "categorical",
     color_mode: str = "grayscale",
-    image_size: Tuple[int, int] = (28, 28),
+    image_size: Tuple[int, int] = (28, 28, 1),
     verbose: bool = True,
     **kwargs,
 ):
-    model_name = model_name + "/" + model_version
+    model_name = model_name + "/" + str(model_version)
     LOGGER.info("Model Selected: ", model_name=model_name)
     color_mode: str = "color" if image_size[2] == 3 else "grayscale"
     target_size: Tuple[int, int] = image_size[:2]

examples/task-plugins/dioptra_custom/feature_squeezing/squeeze_plugin.py

@@ -26,7 +26,7 @@
 import numpy as np
 import structlog
 import tensorflow as tf
-from tensorflow.keras.preprocessing.image import ImageDataGenerator, save_img
+from tensorflow.keras.preprocessing.image import DirectoryIterator, ImageDataGenerator, save_img
 
 from dioptra import pyplugs
 from dioptra.sdk.exceptions import ARTDependencyError, TensorflowDependencyError
@@ -51,19 +51,19 @@
 @require_package("art", exc_type=ARTDependencyError)
 @require_package("tensorflow", exc_type=TensorflowDependencyError)
 def feature_squeeze(
-    data_dir,
-    run_id,
-    model,
-    model_architecture,
-    batch_size,
-    seed,
-    bit_depth,
-    model_version,
-    adv_tar_name,
-    image_size,
-    adv_data_dir,
-    data_flow,
-):
+    data_dir: str,
+    run_id: str,
+    model: str,
+    model_architecture: str,
+    adv_tar_name: str,
+    image_size: Tuple[int,int,int],
+    adv_data_dir: str,
+    data_flow: DirectoryIterator,
+    batch_size: int = 32,
+    seed: int = -1,
+    bit_depth: int = 8,
+    model_version: str = 1,
+) -> None:
     rng = np.random.default_rng(seed if seed >= 0 else None)
 
     if seed < 0:

examples/tensorflow-adversarial-patches/demo-mnist-patches.ipynb

@@ -923,7 +923,7 @@
    "hash": "edee40310913f16e2ca02c1d37887bcb7f07f00399ca119bb7e27de7d632ea99"
   },
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },
@@ -937,7 +937,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.7.10"
+   "version": "3.9.5"
   }
  },
  "nbformat": 4,

examples/tensorflow-mnist-feature-squeezing-legacy/README.md

@@ -0,0 +1,22 @@
+# Tensorflow MNIST Feature Squeezing demo
+
+This example demonstrates using Dioptra to run experiments that investigate the effectiveness of the feature-squeezing defense against a series of evasion attacks against a neural network model.
+The demo can be found in the Jupyter notebook file [demo.ipynb](demo.ipynb).
+
+## Running the example
+
+To prepare your environment for running this example, follow the linked instructions below:
+
+1.  [Create and activate a Python virtual environment and install the necessary dependencies](../README.md#creating-a-virtual-environment)
+2.  [Download the MNIST dataset using the download_data.py script.](../README.md#downloading-datasets)
+3.  [Follow the links in these User Setup instructions](../../README.md#user-setup) to do the following:
+    -   Build the containers
+    -   Use the cookiecutter template to generate the scripts, configuration files, and Docker Compose files you will need to run Dioptra
+4.  [Edit the docker-compose.yml file to mount the data folder in the worker containers](../README.md#mounting-the-data-folder-in-the-worker-containers)
+5.  [Initialize and start Dioptra](https://pages.nist.gov/dioptra/getting-started/running-dioptra.html#initializing-the-deployment)
+6.  [Register the custom task plugins for Dioptra's examples and demos](../README.md#registering-custom-task-plugins)
+7.  [Register the queues for Dioptra's examples and demos](../README.md#registering-queues)
+8.  [Start JupyterLab and open `demo.ipynb`](../README.md#starting-jupyter-lab)
+
+Steps 1–4 and 6–7 only need to be run once.
+**Returning users only need to repeat Steps 5 (if you stopped Dioptra using `docker compose down`) and 8 (if you stopped the `jupyter lab` process)**.