def plot_image_and_reconstruction(X: np.ndarray,
model: keras.models.Model,
idx: Optional[int] = None,
cmap: Optional[str] = 'binary') -> None:
"""Plot an image and its reconstruction from an autoencoder model
Args:
X (np.ndarray): Training data of autoencoder
model (keras.models.Model): Autoencoder
idx (Optional[int], optional): The index of the training sample to plot. Defaults to None, in which case a random index is chosen.
cmap (Optional[str], optional): The image mapping for pyplot. Defaults to 'binary'.
"""
idx = idx or np.random.choice(len(X))
image = X[idx]
[reconstruction] = model.predict(image[np.newaxis, :])
difference = np.abs(image - reconstruction)
fig = plt.figure(figsize=(3 * 1.5, 3))
plt.subplot(1, 3, 1)
plt.imshow(image, cmap=cmap)
plt.subplot(1, 3, 2)
plt.imshow(reconstruction, cmap=cmap)
plt.subplot(1, 3, 3)
plt.imshow(difference, cmap=cmap)
plt.show()
def split_dataset(dataset: Dataset, model: keras.models.Model) -> Split:
dataset_gen = data_flow_from_disk(SOURCE_PATH, dataset, label_indices,
False, BS, SEED, MODEL)
predictions: Predictions = OrderedDict()
for dataset_item, prediction in zip(dataset.keys(),
model.predict(dataset_gen)):
predictions[dataset_item] = prediction
return SPLITTER_FACTORY(predictions, 50)(dataset)
def predict_future_vals(model: keras.models.Model, window_frame: np.array):
future_window = model.predict(window_frame)
future_window_rounded = np.rint(future_window)
future_window_rounded = np.squeeze(future_window_rounded)
return future_window_rounded