def run_mnist(n_train: int = 4000,
n_test: int = 100,
visualize: bool = True,
n_epochs: int = 5,
l_rate: float = 0.7,
profile: bool = False):
Utilities.delete_previous_output("mnist_images")
mnist_features, mnist_labels, mnist_test_features, mnist_test_labels = DataReader.load_mnist(
train_limit=n_train, test_limit=n_test)
som = SOM(mnist=True,
features=mnist_features,
labels=mnist_labels,
test_features=mnist_test_features,
test_labels=mnist_test_labels,
n_epochs=n_epochs,
initial_radius=5,
initial_l_rate=l_rate,
radius_decay_func="pow",
l_rate_decay_func="pow",
n_output_cols=20,
n_output_rows=20,
display_interval=1 if visualize else -1)
pr = cProfile.Profile()
if profile:
pr.enable()
som.run()
if profile:
pr.disable()
pr.print_stats(sort='time')
if visualize:
Utilities.make_gif(mnist=True)
def run_tsm(city: int,
visualize: bool = True,
profile: bool = False,
n_epochs: int = 400,
l_rate: float = 0.39):
Utilities.delete_previous_output("tsm_images")
cities = DataReader.read_tsm_file(city)
norm_cities = cities
features = norm_cities[:, 1:]
# TSM Hyper Params
node_factor = 3
radius_divisor = 1
l_decay = "cur"
r_decay = "pow"
out_size = len(features) * node_factor
init_rad = int(out_size / radius_divisor)
som = SOM(mnist=False,
features=features,
n_epochs=n_epochs,
n_output_rows=1,
n_output_cols=out_size,
initial_radius=init_rad,
initial_l_rate=l_rate,
radius_decay_func=r_decay,
l_rate_decay_func=l_decay,
originals=cities[:, 1:],
display_interval=10 if visualize else -1)
pr = cProfile.Profile()
if profile:
pr.enable()
result = som.run()
if profile:
pr.disable()
pr.print_stats(sort='time')
Utilities.store_tsm_result(case=city,
epochs=n_epochs,
nodes=node_factor,
l_rate=l_rate,
radius=radius_divisor,
l_decay=l_decay,
r_decay=r_decay,
result=result)
Utilities.make_gif(mnist=False) if visualize else NoOp
return result
