Adding DataViz Driver

Callback object contains intermediate results from evaluator function and
will output results if user sends a keyboard interrupt.

.gitignore

@@ -113,3 +113,4 @@ venv.bak/
 *.model
 *.csv
 *.txt
+!requirements.txt

eyeballer/augmentation.py

@@ -0,0 +1,18 @@
+import numpy as np
+
+from Augmentor.Operations import Operation
+from keras.applications.mobilenet import preprocess_input
+
+
+class EyeballerAugmentation(Operation):
+    def __init__(self, probability=1):
+        Operation.__init__(self, probability)
+
+    # Class must implement the perform_operation method
+    def perform_operation(self, images):
+        return_list = []
+        for image in images:
+            image_array = np.array(image).astype('uint8')
+            image_array = preprocess_input(image_array)
+            return_list.append(image_array)
+        return return_list

eyeballer/conv_filter_visualization.py

@@ -0,0 +1,263 @@
+"""
+#from https://raw.githubusercontent.com/keras-team/keras/master/examples/conv_filter_visualization.py
+
+#Visualization of the filters of VGG16, via gradient ascent in input space.
+
+This script can run on CPU in a few minutes.
+
+Results example: ![Visualization](http://i.imgur.com/4nj4KjN.jpg)
+"""
+from __future__ import print_function
+
+import time
+import numpy as np
+from PIL import Image as pil_image
+from keras.preprocessing.image import save_img
+from keras import layers
+from keras.applications import vgg16
+from keras import backend as K
+
+
+def normalize(x):
+    """utility function to normalize a tensor.
+
+    # Arguments
+        x: An input tensor.
+
+    # Returns
+        The normalized input tensor.
+    """
+    return x / (K.sqrt(K.mean(K.square(x))) + K.epsilon())
+
+
+def deprocess_image(x):
+    """utility function to convert a float array into a valid uint8 image.
+
+    # Arguments
+        x: A numpy-array representing the generated image.
+
+    # Returns
+        A processed numpy-array, which could be used in e.g. imshow.
+    """
+    # normalize tensor: center on 0., ensure std is 0.25
+    x -= x.mean()
+    x /= (x.std() + K.epsilon())
+    x *= 0.25
+
+    # clip to [0, 1]
+    x += 0.5
+    x = np.clip(x, 0, 1)
+
+    # convert to RGB array
+    x *= 255
+    if K.image_data_format() == 'channels_first':
+        x = x.transpose((1, 2, 0))
+    x = np.clip(x, 0, 255).astype('uint8')
+    return x
+
+
+def process_image(x, former):
+    """utility function to convert a valid uint8 image back into a float array.
+       Reverses `deprocess_image`.
+
+    # Arguments
+        x: A numpy-array, which could be used in e.g. imshow.
+        former: The former numpy-array.
+                Need to determine the former mean and variance.
+
+    # Returns
+        A processed numpy-array representing the generated image.
+    """
+    if K.image_data_format() == 'channels_first':
+        x = x.transpose((2, 0, 1))
+    return (x / 255 - 0.5) * 4 * former.std() + former.mean()
+
+
+def visualize_layer(model,
+                    layer_name,
+                    step=1.,
+                    epochs=15,
+                    upscaling_steps=9,
+                    upscaling_factor=1.2,
+                    output_dim=(412, 412),
+                    filter_range=(0, None)):
+    """Visualizes the most relevant filters of one conv-layer in a certain model.
+
+    # Arguments
+        model: The model containing layer_name.
+        layer_name: The name of the layer to be visualized.
+                    Has to be a part of model.
+        step: step size for gradient ascent.
+        epochs: Number of iterations for gradient ascent.
+        upscaling_steps: Number of upscaling steps.
+                         Starting image is in this case (80, 80).
+        upscaling_factor: Factor to which to slowly upgrade
+                          the image towards output_dim.
+        output_dim: [img_width, img_height] The output image dimensions.
+        filter_range: Tupel[lower, upper]
+                      Determines the to be computed filter numbers.
+                      If the second value is `None`,
+                      the last filter will be inferred as the upper boundary.
+    """
+
+    def _generate_filter_image(input_img,
+                               layer_output,
+                               filter_index):
+        """Generates image for one particular filter.
+
+        # Arguments
+            input_img: The input-image Tensor.
+            layer_output: The output-image Tensor.
+            filter_index: The to be processed filter number.
+                          Assumed to be valid.
+
+        #Returns
+            Either None if no image could be generated.
+            or a tuple of the image (array) itself and the last loss.
+        """
+        s_time = time.time()
+
+        # we build a loss function that maximizes the activation
+        # of the nth filter of the layer considered
+        if K.image_data_format() == 'channels_first':
+            loss = K.mean(layer_output[:, filter_index, :, :])
+        else:
+            loss = K.mean(layer_output[:, :, :, filter_index])
+
+        # we compute the gradient of the input picture wrt this loss
+        grads = K.gradients(loss, input_img)[0]
+
+        # normalization trick: we normalize the gradient
+        grads = normalize(grads)
+
+        # this function returns the loss and grads given the input picture
+        iterate = K.function([input_img], [loss, grads])
+
+        # we start from a gray image with some random noise
+        intermediate_dim = tuple(
+            int(x / (upscaling_factor ** upscaling_steps)) for x in output_dim)
+        if K.image_data_format() == 'channels_first':
+            input_img_data = np.random.random(
+                (1, 3, intermediate_dim[0], intermediate_dim[1]))
+        else:
+            input_img_data = np.random.random(
+                (1, intermediate_dim[0], intermediate_dim[1], 3))
+        input_img_data = (input_img_data - 0.5) * 20 + 128
+
+        # Slowly upscaling towards the original size prevents
+        # a dominating high-frequency of the to visualized structure
+        # as it would occur if we directly compute the 412d-image.
+        # Behaves as a better starting point for each following dimension
+        # and therefore avoids poor local minima
+        for up in reversed(range(upscaling_steps)):
+            # we run gradient ascent for e.g. 20 steps
+            for _ in range(epochs):
+                loss_value, grads_value = iterate([input_img_data])
+                input_img_data += grads_value * step
+
+                # some filters get stuck to 0, we can skip them
+                if loss_value <= K.epsilon():
+                    return None
+
+            # Calulate upscaled dimension
+            intermediate_dim = tuple(
+                int(x / (upscaling_factor ** up)) for x in output_dim)
+            # Upscale
+            img = deprocess_image(input_img_data[0])
+            img = np.array(pil_image.fromarray(img).resize(intermediate_dim,
+                                                           pil_image.BICUBIC))
+            input_img_data = [process_image(img, input_img_data[0])]
+
+        # decode the resulting input image
+        img = deprocess_image(input_img_data[0])
+        e_time = time.time()
+        print('Costs of filter {:3}: {:5.0f} ( {:4.2f}s )'.format(filter_index,
+                                                                  loss_value,
+                                                                  e_time - s_time))
+        return img, loss_value
+
+    def _draw_filters(filters, n=None):
+        """Draw the best filters in a nxn grid.
+
+        # Arguments
+            filters: A List of generated images and their corresponding losses
+                     for each processed filter.
+            n: dimension of the grid.
+               If none, the largest possible square will be used
+        """
+        if n is None:
+            n = int(np.floor(np.sqrt(len(filters))))
+
+        # the filters that have the highest loss are assumed to be better-looking.
+        # we will only keep the top n*n filters.
+        filters.sort(key=lambda x: x[1], reverse=True)
+        filters = filters[:n * n]
+
+        # build a black picture with enough space for
+        # e.g. our 8 x 8 filters of size 412 x 412, with a 5px margin in between
+        MARGIN = 5
+        width = n * output_dim[0] + (n - 1) * MARGIN
+        height = n * output_dim[1] + (n - 1) * MARGIN
+        stitched_filters = np.zeros((width, height, 3), dtype='uint8')
+
+        # fill the picture with our saved filters
+        for i in range(n):
+            for j in range(n):
+                img, _ = filters[i * n + j]
+                width_margin = (output_dim[0] + MARGIN) * i
+                height_margin = (output_dim[1] + MARGIN) * j
+                stitched_filters[
+                    width_margin: width_margin + output_dim[0],
+                    height_margin: height_margin + output_dim[1], :] = img
+
+        # save the result to disk
+        save_img('viz_{0:}_{1:}x{1:}.png'.format(layer_name, n), stitched_filters)
+
+    # this is the placeholder for the input images
+    assert len(model.inputs) == 1
+    input_img = model.inputs[0]
+
+    # get the symbolic outputs of each "key" layer (we gave them unique names).
+    layer_dict = dict([(layer.name, layer) for layer in model.layers[1:]])
+
+    output_layer = layer_dict[layer_name]
+    assert isinstance(output_layer, layers.Conv2D)
+
+    # Compute to be processed filter range
+    import pdb
+    pdb.set_trace()
+    filter_lower = filter_range[0]
+    filter_upper = (filter_range[1]
+                    if filter_range[1] is not None
+                    else len(output_layer.get_weights()[0][0][0][0]))
+                    # BF: Using size instead of [0] to account for n-dim array
+    assert(filter_lower >= 0
+           and filter_upper <= len(output_layer.get_weights()[0][0][0][0])
+           and filter_upper > filter_lower)
+    print('Compute filters {:} to {:}'.format(filter_lower, filter_upper))
+
+    # iterate through each filter and generate its corresponding image
+    processed_filters = []
+    for f in range(filter_lower, filter_upper):
+        img_loss = _generate_filter_image(input_img, output_layer.output, f)
+
+        if img_loss is not None:
+            processed_filters.append(img_loss)
+
+    print('{} filter processed.'.format(len(processed_filters)))
+    # Finally draw and store the best filters to disk
+    _draw_filters(processed_filters)
+
+
+if __name__ == '__main__':
+    # the name of the layer we want to visualize
+    # (see model definition at keras/applications/vgg16.py)
+    LAYER_NAME = 'block5_conv1'
+
+    # build the VGG16 network with ImageNet weights
+    vgg = vgg16.VGG16(weights='imagenet', include_top=False)
+    print('Model loaded.')
+    vgg.summary()
+
+    # example function call
+    visualize_layer(vgg, LAYER_NAME)

eyeballer/deep_dream.py

@@ -0,0 +1,182 @@
+'''
+#Deep Dreaming in Keras.
+
+Run the script with:
+```python
+python deep_dream.py path_to_your_base_image.jpg prefix_for_results
+```
+e.g.:
+```python
+python deep_dream.py img/mypic.jpg results/dream
+```
+'''
+from __future__ import print_function
+
+from keras.preprocessing.image import load_img, save_img, img_to_array
+import numpy as np
+import scipy
+import argparse
+
+from keras.applications import inception_v3
+from keras import backend as K
+
+fetch_loss_and_grads = None
+# These are the names of the layers
+# for which we try to maximize activation,
+# as well as their weight in the final loss
+# we try to maximize.
+# You can tweak these setting to obtain new visual effects.
+settings = {
+    'features': {
+        'conv1': 1.,
+        #'mixed2': 0.2,
+        #'mixed3': 0.5,
+        #'mixed4': 2.,
+        #'mixed5': 1.5,
+    },
+}
+
+
+def preprocess_image(image_path):
+    # Util function to open, resize and format pictures
+    # into appropriate tensors.
+    img = load_img(image_path)
+    img = img_to_array(img)
+    img = np.expand_dims(img, axis=0)
+    img = inception_v3.preprocess_input(img)
+    return img
+
+
+def deprocess_image(x):
+    # Util function to convert a tensor into a valid image.
+    if K.image_data_format() == 'channels_first':
+        x = x.reshape((3, x.shape[2], x.shape[3]))
+        x = x.transpose((1, 2, 0))
+    else:
+        x = x.reshape((x.shape[1], x.shape[2], 3))
+    x /= 2.
+    x += 0.5
+    x *= 255.
+    x = np.clip(x, 0, 255).astype('uint8')
+    return x
+
+
+
+def eval_loss_and_grads(x):
+    outs = fetch_loss_and_grads([x])
+    loss_value = outs[0]
+    grad_values = outs[1]
+    return loss_value, grad_values
+
+
+def resize_img(img, size):
+    img = np.copy(img)
+    if K.image_data_format() == 'channels_first':
+        factors = (1, 1,
+                   float(size[0]) / img.shape[2],
+                   float(size[1]) / img.shape[3])
+    else:
+        factors = (1,
+                   float(size[0]) / img.shape[1],
+                   float(size[1]) / img.shape[2],
+                   1)
+    return scipy.ndimage.zoom(img, factors, order=1)
+
+
+def gradient_ascent(x, iterations, step, max_loss=None):
+    for i in range(iterations):
+        loss_value, grad_values = eval_loss_and_grads(x)
+        if max_loss is not None and loss_value > max_loss:
+            break
+        print('..Loss value at', i, ':', loss_value)
+        x += step * grad_values
+    return x
+
+
+"""Process:
+
+- Load the original image.
+- Define a number of processing scales (i.e. image shapes),
+    from smallest to largest.
+- Resize the original image to the smallest scale.
+- For every scale, starting with the smallest (i.e. current one):
+    - Run gradient ascent
+    - Upscale image to the next scale
+    - Reinject the detail that was lost at upscaling time
+- Stop when we are back to the original size.
+
+To obtain the detail lost during upscaling, we simply
+take the original image, shrink it down, upscale it,
+and compare the result to the (resized) original image.
+"""
+def process(model, base_image_path, result_prefix):
+    global settings
+    global fetch_loss_and_grads
+    K.set_learning_phase(0)
+
+    dream = model.input
+    print('Model loaded.')
+
+    # Get the symbolic outputs of each "key" layer (we gave them unique names).
+    layer_dict = dict([(layer.name, layer) for layer in model.layers])
+
+    # Define the loss.
+    loss = K.variable(0.)
+    for layer_name in settings['features']:
+        # Add the L2 norm of the features of a layer to the loss.
+        if layer_name not in layer_dict:
+            raise ValueError('Layer ' + layer_name + ' not found in model.')
+        coeff = settings['features'][layer_name]
+        x = layer_dict[layer_name].output
+        # We avoid border artifacts by only involving non-border pixels in the loss.
+        scaling = K.prod(K.cast(K.shape(x), 'float32'))
+        if K.image_data_format() == 'channels_first':
+            loss += coeff * K.sum(K.square(x[:, :, 2: -2, 2: -2])) / scaling
+        else:
+            loss += coeff * K.sum(K.square(x[:, 2: -2, 2: -2, :])) / scaling
+
+    # Compute the gradients of the dream wrt the loss.
+    grads = K.gradients(loss, dream)[0]
+    # Normalize gradients.
+    grads /= K.maximum(K.mean(K.abs(grads)), K.epsilon())
+
+    # Set up function to retrieve the value
+    # of the loss and gradients given an input image.
+    outputs = [loss, grads]
+    fetch_loss_and_grads = K.function([dream], outputs)
+
+    # Playing with these hyperparameters will also allow you to achieve new effects
+    step = 0.01  # Gradient ascent step size
+    num_octave = 3  # Number of scales at which to run gradient ascent
+    octave_scale = 1.4  # Size ratio between scales
+    iterations = 20  # Number of ascent steps per scale
+    max_loss = 10.
+
+    img = preprocess_image(base_image_path)
+    if K.image_data_format() == 'channels_first':
+        original_shape = img.shape[2:]
+    else:
+        original_shape = img.shape[1:3]
+    successive_shapes = [original_shape]
+    for i in range(1, num_octave):
+        shape = tuple([int(dim / (octave_scale ** i)) for dim in original_shape])
+        successive_shapes.append(shape)
+    successive_shapes = successive_shapes[::-1]
+    original_img = np.copy(img)
+    shrunk_original_img = resize_img(img, successive_shapes[0])
+
+    for shape in successive_shapes:
+        print('Processing image shape', shape)
+        img = resize_img(img, shape)
+        img = gradient_ascent(img,
+                            iterations=iterations,
+                            step=step,
+                            max_loss=max_loss)
+        upscaled_shrunk_original_img = resize_img(shrunk_original_img, shape)
+        same_size_original = resize_img(original_img, shape)
+        lost_detail = same_size_original - upscaled_shrunk_original_img
+
+        img += lost_detail
+        shrunk_original_img = resize_img(original_img, shape)
+
+    save_img(result_prefix + '.png', deprocess_image(np.copy(img)))

eyeballer/eyeballer.py

@@ -1,6 +1,11 @@
 #!/usr/bin/env python3
 import click
-from model import EyeballModel
+import csv
+from conv_filter_visualization import visualize_layer
+from keras import layers
+import deep_dream
+
+from model import EyeballModel, DATA_LABELS
 
 
 @click.group(invoke_without_command=True)
@@ -31,24 +36,33 @@ def train(ctx, graphs, batchsize, epochs):
 
 
 @cli.command()
-@click.argument('screenshot', nargs=-1)  # Nargs=-1 means you can pass as many in as you want
+@click.argument('screenshot')
 @click.pass_context
 def predict(ctx, screenshot):
     model = EyeballModel(**ctx.obj['model_kwargs'])
-    for fn in screenshot:
-        model.predict(fn)
+    results = model.predict(screenshot)
+    if not results:
+        print("Error: Input file does not exist")
+    if len(results) == 1:
+        print(results)
+    else:
+        with open("results.csv", "w", newline="") as csvfile:
+            fieldnames = ["filename", "custom404", "login", "homepage", "oldlooking"]
+            labelwriter = csv.DictWriter(csvfile, fieldnames=fieldnames)
+            labelwriter.writeheader()
+            labelwriter.writerows(results)
+        print("Output written to results.csv")
 
 
 def pretty_print_evaluation(results):
-    """Print a human-readable summary of the evaulation results"""
-    print("Custom404 Precision Score: " + str(round(results["custom404"]["precision"] * 100, 2)) + "%")
-    print("Custom404 Recall Score: " + str(round(results["custom404"]["recall"] * 100, 2)) + "%")
-    print("Login Precision Score: " + str(round(results["login"]["precision"] * 100, 2)) + "%")
-    print("Login Recall Score: " + str(round(results["login"]["recall"] * 100, 2)) + "%")
-    print("Homepage Precision Score: " + str(round(results["homepage"]["precision"] * 100, 2)) + "%")
-    print("Homepage Recall Score: " + str(round(results["homepage"]["recall"] * 100, 2)) + "%")
-    print("Overall Binary Accuracy: " + str(round(results["total_binary_accuracy"] * 100, 2)) + "%")
-    print("All or nothing accuracy: " + str(round(results["all_or_nothing_accuracy"] * 100, 2)) + "%")
+    """Print a human-readable summary of the evaluation"""
+    # We use 4.2% to handle all the way from "  0.00%" (7chars) to "100.00%" (7chars)
+    for label in DATA_LABELS:
+        print(f"{label} Precision Score: {results[label]['precision']:4.2%}")
+        print(f"{label} Recall Score: {results[label]['recall']:4.2%}")
+    print(f"Overall Binary Accuracy: {results['total_binary_accuracy']:4.2%}")
+    print(f"All or nothing accuracy: {results['all_or_nothing_accuracy']:4.2%}")
+    print(f"Top 10 worst predictions: {results['top_10_worst'][1]}")
 
 
 @cli.command()
@@ -60,5 +74,29 @@ def evaluate(ctx, threshold):
     pretty_print_evaluation(results)
 
 
+@cli.command()
+@click.argument('layer', default=None, required=False)
+@click.pass_context
+def visualize(ctx, layer):
+    model = EyeballModel(**ctx.obj['model_kwargs']).model
+    if not layer:
+        layer_dict = dict([(layer.name, layer) for layer in model.layers[1:]])
+        print("Layers:")
+        for k in layer_dict.keys():
+            if isinstance(layer_dict[k], layers.Conv2D):
+                print("\t" + k)
+    else:
+        visualize_layer(model, layer)
+
+@cli.command()
+@click.option('--base-image-path', type=click.Path(),
+                    help='Path to the image to transform.')
+@click.option('--result-prefix', type=click.Path(),
+                    help='Prefix for the saved results.')
+@click.pass_context
+def deepdream(ctx, base_image_path, result_prefix):
+    model = EyeballModel(**ctx.obj['model_kwargs']).model
+    deep_dream.process(model, base_image_path, result_prefix)
+
 if __name__ == '__main__':
     cli()

eyeballer/model.py

@@ -1,10 +1,14 @@
 import os
 import random
-import time
 
+# Prevent Tkinter Dependency
+import matplotlib
+matplotlib.use('agg')  # noqa: E402
 import matplotlib.pyplot as plt
+
 import numpy as np
 import pandas as pd
+import Augmentor
 
 from keras import regularizers
 from keras.applications import MobileNet
@@ -16,9 +20,23 @@ from keras.models import Model
 from keras.optimizers import Adam
 from keras.preprocessing import image
 from keras.preprocessing.image import ImageDataGenerator
-from sklearn.metrics import classification_report, accuracy_score, jaccard_similarity_score
+from sklearn.metrics import classification_report, accuracy_score, hamming_loss
+from augmentation import EyeballerAugmentation
+from keras.callbacks import Callback
 
-DATA_LABELS = ["custom404", "login", "homepage"]
+DATA_LABELS = ["custom404", "login", "homepage", "oldlooking"]
+
+
+class PredictionManager(Callback):
+    def __init__(self):
+        super(PredictionManager, self).__init__()
+        self.predictions = []
+
+    def getPredictions(self):
+        return np.array(self.predictions)
+
+    def on_predict_batch_end(self, batch, logs=None):
+        self.predictions.append(logs['outputs'][0].tolist()[0])
 
 
 class EyeballModel:
@@ -26,7 +44,8 @@ class EyeballModel:
 
     Contains high-level functions for training, evaluating, and predicting.
     """
-    checkpoint_file = "weights-snapshot.h5"
+    graphs_directory = "graphs/"
+    checkpoint_file = "weights.h5"
     image_dir = "images_224x224/"
     image_width, image_height = 224, 224
     input_shape = (image_width, image_height, 3)
@@ -47,11 +66,11 @@ class EyeballModel:
         x = Dropout(0.5)(x)
         x = Dense(128, activation="relu", name="HiddenLayer2", kernel_regularizer=regularizers.l2(0.01))(x)
         x = Dropout(0.2)(x)
-        output_layer = Dense(3, activation="sigmoid", name="OutputLayer")(x)
+        output_layer = Dense(len(DATA_LABELS), activation="sigmoid", name="OutputLayer")(x)
         self.model = Model(inputs=base_model.input, outputs=output_layer)
 
         # for layer in base_model.layers:
-        #    layer.trainable = False
+        #     layer.trainable = False
 
         adam = Adam(lr=0.0005)
         self.model.compile(optimizer=adam, loss="binary_crossentropy", metrics=["accuracy"])
@@ -81,6 +100,20 @@ class EyeballModel:
         else:
             print("No model loaded from file")
 
+        # Data augmentation
+        augmentor = Augmentor.Pipeline()
+        augmentor.set_seed(self.seed)
+        augmentor.zoom(probability=0.75, min_factor=0.8, max_factor=1.2)
+        augmentor.random_color(probability=0.75, min_factor=0.5, max_factor=1.0)
+        augmentor.random_contrast(probability=0.75, min_factor=0.8, max_factor=1.0)
+        augmentor.random_brightness(probability=0.75, min_factor=0.8, max_factor=1.2)
+        augmentor.random_erasing(probability=0.75, rectangle_area=0.15)
+
+        # Finalizes the augmentation with a custom operation to prepare the image for the specific pretrained model we're using
+        training_augmentation = EyeballerAugmentation()
+        augmentor.add_operation(training_augmentation)
+        self.preprocess_training_function = augmentor.keras_preprocess_func()
+
     def train(self, epochs=20, batch_size=32, print_graphs=False):
         """Train the model, making a new weights file at each successfull checkpoint. You'll probably need a GPU for this to realistically run.
 
@@ -91,15 +124,10 @@ class EyeballModel:
         """
         print("Training with seed: " + str(self.seed))
 
-        # Data augmentation
         data_generator = ImageDataGenerator(
-            preprocessing_function=preprocess_input,
-            shear_range=0.0,
-            zoom_range=0.2,
-            brightness_range=(0.7, 1.3),
-            samplewise_center=True,
+            preprocessing_function=self.preprocess_training_function,
             validation_split=0.2,
-            horizontal_flip=False)
+            samplewise_center=True)
 
         training_generator = data_generator.flow_from_dataframe(
             self.training_labels,
@@ -138,6 +166,8 @@ class EyeballModel:
             verbose=1)
 
         if print_graphs:
+            if not os.path.exists(self.graphs_directory):
+                os.makedirs(self.graphs_directory)
             # Plot training & validation accuracy values
             plt.plot(history.history['acc'])
             plt.plot(history.history['val_acc'])
@@ -145,7 +175,7 @@ class EyeballModel:
             plt.ylabel('Accuracy')
             plt.xlabel('Epoch')
             plt.legend(['Train', 'Validation'], loc='upper left')
-            plt.savefig("accuracy.png")
+            plt.savefig(self.graphs_directory + "accuracy.png")
             plt.clf()
             plt.cla()
             plt.close()
@@ -157,28 +187,42 @@ class EyeballModel:
             plt.ylabel('Loss')
             plt.xlabel('Epoch')
             plt.legend(['Train', 'Validation'], loc='upper left')
-            plt.savefig("loss.png")
+            plt.savefig(self.graphs_directory + "loss.png")
 
-    def predict(self, filename):
+    def predict(self, path):
         """Predict the labels for a single file
 
         Keyword arguments:
-        filename -- The path to the file that we'll be evaluating
+        path -- The path to the file that we'll be evaluating
         """
-        # Load the image into memory
-        img = image.load_img(filename, target_size=(self.image_width, self.image_height))
-        img = image.img_to_array(img)
-        img = np.expand_dims(img, axis=0)
-        img = preprocess_input(img)
+        # Is this a single file, or a directory?
+        screenshots = []
+        if os.path.isfile(path):
+            screenshots = [path]
+        elif os.path.isdir(path):
+            screenshots = os.listdir(path)
+            screenshots = [os.path.join(path, s) for s in screenshots]
+        else:
+            return None
 
-        time_start = time.time()
-        prediction = self.model.predict(img, batch_size=1)
-        time_end = time.time()
-        print("Predictions:")
-        print("\tCustom 404: " + str(round(prediction[0][0] * 100, 2)) + "%")
-        print("\tLogin Page: " + str(round(prediction[0][1] * 100, 2)) + "%")
-        print("\tHome Page: " + str(round(prediction[0][2] * 100, 2)) + "%")
-        print("Prediction Took (seconds): ", time_end - time_start)
+        results = []
+        for screenshot in screenshots:
+            # Load the image into memory
+            img = image.load_img(screenshot, target_size=(self.image_width, self.image_height))
+            img = image.img_to_array(img)
+            img = np.expand_dims(img, axis=0)
+            img = preprocess_input(img)
+
+            prediction = self.model.predict(img, batch_size=1)
+            result = dict()
+            result["filename"] = screenshot
+            result["custom404"] = prediction[0][0]
+            result["login"] = prediction[0][1]
+            result["homepage"] = prediction[0][2]
+            result["oldlooking"] = prediction[0][3]
+            results.append(result)
+
+        return results
 
     def evaluate(self, threshold=0.5):
         """Evaluate performance against the persistent evaluation data set
@@ -203,13 +247,9 @@ class EyeballModel:
             print("Using validation set...")
             # Data augmentation
             data_generator = ImageDataGenerator(
-                preprocessing_function=preprocess_input,
-                shear_range=0.0,
-                zoom_range=0.2,
-                brightness_range=(0.7, 1.3),
+                preprocessing_function=self.preprocess_training_function,
                 samplewise_center=True,
-                validation_split=0.2,
-                horizontal_flip=False)
+                validation_split=0.2)
             evaluation_generator = data_generator.flow_from_dataframe(
                 self.training_labels,
                 directory=self.image_dir,
@@ -224,14 +264,77 @@ class EyeballModel:
         else:
             print("Using evaluation set...")
 
-        predictions = self.model.predict_generator(evaluation_generator, verbose=1, steps=len(evaluation_generator))
+        prediction_manager = PredictionManager()
+
+        try:
+            self.model.predict_generator(
+                evaluation_generator,
+                callbacks=[prediction_manager],
+                verbose=1,
+                steps=len(evaluation_generator))
+        except KeyboardInterrupt:
+            print('\r\nJob interrupted (Ctrl+C): Outputting completed results')
+
+        predictions = prediction_manager.getPredictions()
+
+        # Truncates evaluation generator if test was interrupted
+        evaluation_generator._data = evaluation_generator.data[:predictions.shape[0]]
+
+        self._save_prediction_histograms(predictions)
         predictions = predictions > threshold
         ground_truth = evaluation_generator.data
-        
         stats = classification_report(ground_truth, predictions, target_names=DATA_LABELS, output_dict=True)
-        stats["total_binary_accuracy"] = jaccard_similarity_score(ground_truth, predictions)
+        stats["total_binary_accuracy"] = 1 - hamming_loss(ground_truth, predictions)
         stats["all_or_nothing_accuracy"] = accuracy_score(ground_truth, predictions)
+        stats["top_10_best"] = self._top_images(evaluation_generator, predictions, best=True)
+        stats["top_10_worst"] = self._top_images(evaluation_generator, predictions, best=False)
         return stats
 
-    def get_data_column(self, data_slice, data):
-        return np.reshape(np.array(data)[data_slice], len(data)).tolist()
+    def _top_images(self, labels, predictions, top_k=10, best=False):
+        """Collect top-k best or top-k worst predicted images
+
+        Keyword arguments:
+        labels -- The keras generator that contains the filenames and data labels
+        predictions -- The numpy array of predictions
+        top_k -- Top k elements
+        best -- True/False. Calculate either the best or worst images
+
+        :Return -- Tuple of top-k indicies and top-k filenames
+        """
+        true_labels = np.array(labels.data).astype(float)
+        predictions = predictions.astype(float)
+
+        differences = np.abs(predictions - true_labels).sum(axis=1)
+        indicies = np.argsort(differences, axis=0)
+
+        top_file_list = []
+
+        if not best:
+            # Reverse numpy array
+            indicies = np.flipud(indicies)
+
+        for i in indicies[:top_k]:
+            top_file_list.append(labels.filenames[i])
+
+        return indicies[:top_k], top_file_list
+
+    def _save_prediction_histograms(self, predictions, buckets=50):
+        """Saves a series of histogram screenshots
+
+            Keyword arguments:
+            predictions -- The numpy array of predicted labels, ranging from 0->1.
+            buckets -- The number of buckets to divide the data into. Default: 50
+            :Returns -- Nothing"
+        """
+        figure, axes = plt.subplots(nrows=len(DATA_LABELS))
+        for i, label in enumerate(DATA_LABELS):
+            axes[i].hist(predictions[:, i], buckets, alpha=.75)
+            axes[i].set_xlabel("Prediction")
+            axes[i].set_ylabel("Counts of Predictions")
+            axes[i].set_title(label)
+            axes[i].grid(True)
+        figure.set_size_inches(5, 3*len(DATA_LABELS))
+        figure.tight_layout()
+        if not os.path.exists(self.graphs_directory):
+            os.makedirs(self.graphs_directory)
+        plt.savefig(self.graphs_directory + "label_histograms.png")

requirements.txt

@@ -0,0 +1,9 @@
+Augmentor
+click
+matplotlib
+numpy
+pandas
+pillow
+sklearn
+tensorflow
+git+https://github.com/keras-team/keras/

setup.py

@@ -1,22 +0,0 @@
-from setuptools import find_packages, setup
-
-setup(
-    name='eyeballer',
-    version='0.0.1',
-    packages=find_packages(exclude=["tests"]),
-    url='https://github.com/BishopFox/eyeballer',
-    license='GNU General Public License v3.0',
-    author='Daniel Petro',
-    author_email='',
-    description='',
-    install_requires=[
-        'keras',
-        'sklearn',
-        'numpy',
-        'pandas',
-        'matplotlib',
-        'pillow', 'click'
-    ],
-    test_requires=[],
-    python_requires='>=3.6',    # We use some python 3.6+ only features, like ordered dicts by default and f-strings
-)

utils/labelbox_to_labels.py

@@ -0,0 +1,37 @@
+#!/usr/bin/env python3
+
+import csv
+import random
+import json
+
+with open("newlabels.csv", "w", newline="") as csvfile:
+    # Open the old labels file
+    with open("labelbox.csv", newline="") as oldfile:
+        # Get the header labels
+        csvreader = csv.DictReader(oldfile)
+        fieldnames = next(csvreader)
+        labelwriter = csv.DictWriter(csvfile, fieldnames=["filename", "login", "custom404", "homepage", "oldlooking", "evaluation"])
+        labelwriter.writeheader()
+
+        # Loop through the file
+        rows = []
+        for row in csvreader:
+            filename = row["External ID"]
+            print(row["Label"])
+            labelstring = row["Label"]
+            if row["Label"] == "Skip":
+                labelstring = '{"imageclassification":[]}'
+            labels = json.loads(labelstring)
+
+            newrow = dict()
+            newrow["filename"] = filename
+            newrow["oldlooking"] = True
+            newrow["login"] = "loginpage" in labels["imageclassification"]
+            newrow["homepage"] = "homepage" in labels["imageclassification"]
+            newrow["custom404"] = "custom404"in labels["imageclassification"]
+
+            newrow["evaluation"] = random.random() > 0.8
+
+            rows.append(newrow)
+        labelwriter.writerows(rows)
+print("Made new labels file: newlabels.csv")