f8a5a458ff6a35e91ae9d705160d73cdcc8adf97,test.py,,,#,13

Before Change

// get adversarial examples
ADV_PATH = os.path.join(DATA_PATH, "adversarial", "mnist", "fgsm", "cnn", "relu", "")

for eps in range(1, 11):
    X = np.load(ADV_PATH + "eps%.2f_train.npy" % (eps/10))

    scores = model.evaluate(X, Y_train, verbose=args.verbose)
    v_print("\naccuracy on train adversarial with  %.2f eps: %.2f%%" % (eps/10, scores[1] * 100))

    X = np.load(ADV_PATH + "eps%.2f_test.npy" % (eps/10))

    scores = model.evaluate(X, Y_test, verbose=args.verbose)
    v_print("\naccuracy on test adversarial with %.2f eps: %.2f%%" % (eps/10, scores[1] * 100))

After Change

model = cnn.load_model(MODEL_PATH, "best-weights.h5")

// ------------------------------------------------------------------------------------------------------- TEST
results = {}

// get MNIST
(X_train, Y_train), (X_test, Y_test) = load_mnist()

// Test on true train instances
scores = model.evaluate(X_train, Y_train, verbose=args.verbose)
v_print("\naccuracy on train: %.2f%%" % (scores[1] * 100))
results["train_accuracy"] = scores[1] * 100

// Test on true test instances
scores = model.evaluate(X_test, Y_test, verbose=args.verbose)
v_print("\naccuracy on test: %.2f%%" % (scores[1] * 100))
results["test_accuracy"] = scores[1] * 10

// get adversarial examples
ADV_PATH = os.path.join(DATA_PATH, "adversarial", "mnist")

for filepath in get_npy_files(ADV_PATH):

    X = np.load(filepath)
    Y = Y_train if "_train" in filepath else Y_test

    scores = model.evaluate(X, Y, verbose=args.verbose)
    v_print("\naccuracy on %s: %.2f%%" % (filepath, scores[1] * 100))
    results[filepath] = scores[1]*100

with open(os.path.join(MODEL_PATH, "accuracies.json"), "w") as json_file:
    json.dump(results, json_file)

Instances