def run_NAS_EA_FA_V2():
best_valid_acc, best_test_acc, times = NAS_EA_FA_V2()
valid_acc, test_acc = handle_result(best_valid_acc, best_test_acc, times,
MAX_TIME_BUDGET, INTERVAL)
return valid_acc, test_acc
def run_regularized_evolution_algorithm(dataset: str):
best_valid_acc, best_test_acc, times = regularized_evolution_algorithm(
dataset)
valid_acc, test_acc = handle_result(best_valid_acc, best_test_acc, times,
MAX_TIME_BUDGET, INTERVAL)
return valid_acc, test_acc
def run_random_search(dataset: str):
best_valid_acc, best_test_acc, times = random_search(dataset)
valid_acc, test_acc = handle_result(best_valid_acc, best_test_acc, times, MAX_TIME_BUDGET, INTERVAL)
return valid_acc, test_acc
def run_neural_predictor():
best_valid_acc, best_test_acc, times = neural_predictor()
valid_acc, test_acc = handle_result(best_valid_acc, best_test_acc, times,
MAX_TIME_BUDGET, INTERVAL)
return valid_acc, test_acc
coco_names = load_classnames('datasets/coco.names')
num_channels = 3
image_filenames = [r"coco-cat-test.jpg"]
raw_images = [None] * len(image_filenames)
batch = torch.empty(
(len(image_filenames), num_channels, input_width, input_height),
dtype=torch.float32)
for b, image_fn in enumerate(image_filenames):
batch[b, :], raw_images[b] = load_image(image_fn,
(input_height, input_width))
batch = batch.to(device)
with torch.no_grad():
# # This is a hack for the fact that the ops inside the nested sequentials
# # seem to be missing the no_grad context
# model.disable_grad()
predictions = model(batch)
palette = pkl.load(open("palette", "rb"))
batch_predictions = handle_result(predictions, 0.5, 0.4)
print_predictions(image_filenames, coco_names, batch_predictions)
plot_predictions(raw_images, coco_names, batch_predictions, palette)
print(builder)