def main_predict() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_08_2')
model = CNNAugDataRegularized((32, 64, 128, 256), (512, 128), 4, device)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 18, 20, tracker, learning_rate=0.0001)
test_X, imgs_ids = data.get_test()
pred_y = model.predict(test_X)
create_submit(pred_y, imgs_ids, path.join(save_dir, 'submission.csv'))
def main_predict() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_25_1')
model = DropoutCNN((32, 64, 128, 256), (512, 128),
4,
device,
drop_dense_p=0.2,
drop_conv_p=0.2)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 30, 20, tracker, learning_rate=0.0001)
test_X, imgs_ids = data.get_test()
pred_y = model.predict(test_X)
create_submit(pred_y, imgs_ids, path.join(save_dir, 'submission.csv'))
def main() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_05_10_1')
model = VGGStyleNet(4, device)
summary(model.cuda(), (3, 128, 128))
print(model)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 40, 64, tracker, learning_rate=0.00001)
tracker.graphs()
tracker.save('metrics.csv')
test_X, imgs_ids = data.get_test()
pred_y = model.predict(test_X)
create_submit(pred_y, imgs_ids, path.join(save_dir, 'submission.csv'))
def main_train() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_08_1')
model = CNNAugDataRegularized((32, 64, 128, 256), (1024, 128), 4, device)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 60, 20, tracker, learning_rate=0.0001)
tracker.graphs()
tracker.save('metrics.csv')
def main_train() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_26_2')
model = TransferCNN(device)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 20, 20, tracker, learning_rate = 0.0001)
tracker.graphs()
tracker.save('metrics.csv')
def main() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_05_10_2')
batch_size = 64
epochs = 15
model = VGGStyleBNNet(4, device)
summary(model.cuda(), (3, 128, 128))
print(model)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
try:
model.train(data, epochs, batch_size, tracker, learning_rate=0.001)
except KeyboardInterrupt:
print('Training interrupted, writing stats...')
finally:
tracker.graphs()
tracker.save('metrics.csv')
test_X, imgs_ids = data.get_test()
pred_y = model.predict(test_X, batch_size)
create_submit(pred_y, imgs_ids, path.join(save_dir, 'submission.csv'))
def main_train() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_26_1')
conv_filter_nums = (16, 32, 64, 64, 128, 128)
neuron_nums = (512, 128)
model = BigCNN(conv_filter_nums, neuron_nums, device)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 40, 20, tracker, learning_rate=0.0001)
tracker.graphs()
tracker.save('metrics.csv')
def main_train() -> None:
detect_gpu()
device = get_device()
save_dir = path.join('modelling', 'model_2020_04_25_1')
model = DropoutCNN((32, 64, 128, 256), (512, 128),
4,
device,
drop_dense_p=0.2,
drop_conv_p=0.2)
data = FirstAugmentedDataset()
tracker = PerformanceTracker(save_dir)
model.train(data, 40, 20, tracker, learning_rate=0.0001)
tracker.graphs()
tracker.save('metrics.csv')