andrewschreiber

from model.data import mnist_train_test_sets
from model.network import LeNet5
from saliency.vanilla_gradient import save_vanilla_gradient
# Get MNIST dataset, preprocessed
train_images, train_labels, test_images, test_labels = mnist_train_test_sets()

# Load net with 98% acc weights
net = LeNet5(weights_path="15epoch_weights.pkl")

# Generate saliency maps for the first 10 images

andrewschreiber

# Create a saliency map for each data point
for i, image in enumerate(data):
    # Run a forward pass with an image
    output = image
    for l in range(len(network.layers)):
        output = network.layers[l].forward(output)

andrewschreiber

    # Backprop to get gradient
    label_one_hot = labels[i]
    dy = np.array(label_one_hot)
    for l in range(len(network.layers)-1, -1, -1):
        dout = network.layers[l].backward(dy)
        dy = dout

andrewschreiber

# Create a saliency map for each data point
for i, image in enumerate(data):
    # Forward pass on image
    # Note: the activations from this are saved on each layer
    output = image
    for l in range(len(network.layers)):
        output = network.layers[l].forward(output)

    # Backprop to get gradient
    label_one_hot = labels[i]

andrewschreiber

# Create a saliency map for each data point
for i, image in enumerate(data):
    # Forward pass on image
    # Note: the activations are saved on each layer
    output = image
    for l in range(len(network.layers)):
        output = network.layers[l].forward(output)

    # Backprop to get gradient
    label_one_hot = labels[i]