def perform_pipeline_for_plots():
y_length, x_length = Config.slide_window
num_of_nodes = y_length * x_length * 3
autoencoder = Autoencoder(
lr=Config.learning_rate,
momentum=0.0,
adaptive_lr=Config.adaptive_lr,
shape=[num_of_nodes, Config.num_of_hidden_neurons, num_of_nodes])
dataset = ImageAutoencoderDataset(image_path=Config.image_path,
image_size=Config.image_size,
slide_window=Config.slide_window)
total_error_list = train_model(network=autoencoder,
dataset=dataset,
n_epochs=Config.n_epochs,
min_error=Config.min_error,
verbose=False)
compression_rate = calculate_image_compression(
num_of_input_neurons=num_of_nodes,
num_of_chunks=len(dataset),
num_of_hidden_neurons=Config.num_of_hidden_neurons)
return total_error_list, compression_rate
def train_eval_factorial(verbose: bool = False) -> Tuple[float, List[float]]:
"""
Trains jordan model on factorial sequence and evaluates it.
Sequence example: [1, 1, 2, 6, 24, 120, 720, ...]
:param verbose: if verbose draws plots
:return: accuracy og the model
"""
dataset = FactorialDataset(
number_of_precalculated_values=Config.num_of_precalculated_values,
number_of_input_elements=Config.num_of_input_elements)
in_features = dataset.number_of_input_elements
out_features = 1
network = Jordan(
lr=Config.learning_rate,
momentum=Config.momentum,
shape=[in_features, Config.num_of_hidden_neurons, out_features])
errors_list = train_model(network=network,
dataset=dataset,
n_epochs=Config.num_epochs)
accuracy = eval_model(network=network, dataset=dataset)
if verbose:
draw_error_plot(errors_list=errors_list,
title='Jordan model on factorial sequence errors')
return accuracy, errors_list
def perform_pipeline():
y_length, x_length = Config.slide_window
num_of_nodes = y_length * x_length * 3
autoencoder = Autoencoder(
lr=Config.learning_rate,
momentum=0.0,
adaptive_lr=Config.adaptive_lr,
shape=[num_of_nodes, Config.num_of_hidden_neurons, num_of_nodes])
dataset = ImageAutoencoderDataset(image_path=Config.image_path,
image_size=Config.image_size,
slide_window=Config.slide_window)
total_error_list = train_model(network=autoencoder,
dataset=dataset,
n_epochs=Config.n_epochs)
res_image = np.zeros(dataset.image.shape)
true_image = dataset.image
for idx, (input_image_flatten, true_image_flatten) in enumerate(dataset):
result = autoencoder.propagate_forward(x=input_image_flatten)
# result = sigmoid(result)
result = result.reshape((*Config.slide_window, 3))
res_image = input_part_into_res_image(res_image=res_image,
image_chunk=result,
idx=idx,
slide_window=Config.slide_window)
res_image = np.uint8(np.clip(res_image * 255, a_min=0, a_max=255))
res_image = cv2.cvtColor(res_image, cv2.COLOR_BGR2RGB)
true_image = cv2.cvtColor(true_image, cv2.COLOR_BGR2RGB)
compression_rate = calculate_image_compression(
num_of_input_neurons=num_of_nodes,
num_of_chunks=len(dataset),
num_of_hidden_neurons=Config.num_of_hidden_neurons)
return res_image, true_image, total_error_list, compression_rate