def main(args):
start = time.time()
set_seed(args)
print("start in {}".format(start))
# load data and preprocess
print("loading tokenizer")
tokenizer = select_tokenizer(args)
model, Processor = select_task(args)
if args.mission == 'train':
print("loading train set")
processor = Processor(args.dataset_dir, 'train')
processor.load_swtc()
train_dataloader = processor.make_dataloader(tokenizer,
args.batch_size, False,
128)
print('loading dev set')
processor = Processor(args.dataset_dir, 'dev')
processor.load_swtc()
deval_dataloader = processor.make_dataloader(tokenizer,
args.batch_size, False,
128)
else:
print(f'loading {args.mission} set')
# file = 'test' if args.mission == 'predict' else 'test'
file = 'target' if args.mission == 'predict' else 'test'
processor = Processor(args.dataset_dir, file)
processor.load_swtc()
test_dataloader = processor.make_dataloader(tokenizer, args.batch_size,
False, 128, False)
# choose model and initalize controller
controller = TextClassification(args)
controller.init(model)
# run task accroading to mission
if args.mission == 'train':
controller.train(train_dataloader,
deval_dataloader,
save_last=args.save_last)
elif args.mission == 'test':
controller.test(test_dataloader)
elif args.mission == "predict":
result, predicts = controller.predict(test_dataloader)
processor.add_predict(predicts)
processor.save_data(args.pred_file_dir)
# with open(args.pred_file_name, 'w', encoding='utf-8') as f:
# f.write(content)
end = time.time()
logger.info("start in {:.0f}, end in {:.0f}".format(start, end))
logger.info("运行时间:%.2f秒" % (end - start))
estimated[estimator_name].append(estimator_value)
value_true = estimated.pop('true')
plt.figure()
plt.plot(parameters, value_true, label='true', ls='--', marker='x', markersize=8)
for estimator_name, estimator_value in estimated.items():
plt.plot(parameters, estimator_value, label=estimator_name)
plt.xlabel(xlabel)
plt.ylabel('Estimated Mutual Information, bits')
plt.title(f"{generator.__name__.lstrip('_mi_')}: len(X)={n_samples}, dim(X)={n_features}")
plt.legend()
plt.savefig(IMAGES_MUTUAL_INFO_DISTRIBUTIONS_DIR / f"{generator.__name__}.png")
# plt.show()
def mi_all_tests(n_samples=10_000, n_features=10):
set_seed(26)
mi_test(_mi_uniform_squared, n_samples=n_samples, n_features=n_features,
xlabel=r'$X \sim $Uniform$(0, x); Y = X^2$')
mi_test(_mi_squared_integers, n_samples=n_samples, n_features=n_features,
xlabel=r'$X \sim $Randint$(0, x); Y = X^2$')
mi_test(_mi_normal_correlated, n_samples=n_samples, n_features=n_features,
xlabel=r'$XY \sim \mathcal{N}(0, \Sigma^\top \Sigma), \Sigma_{ij} \sim $Uniform$(0, x)$')
mi_test(_mi_additive_normal_noise, n_samples=n_samples, n_features=n_features,
xlabel=r'$X \sim \mathcal{N}(0, x^2); Y = X + \epsilon,'
r'\epsilon \sim \mathcal{N}(0,\left(\frac{x}{2}\right)^2$)')
mi_test(_mi_normal_different_location, n_samples=n_samples, n_features=n_features,
xlabel=r'$XY \sim \mathcal{N}(\mu, 10^2), \mu \sim $Uniform$(0, x)$')
Timer.checkpoint()
if __name__ == '__main__':
estimated_test = MutualInfoTest(x=x,
y=y,
mi_true=mi_true,
verbose=False).run_all()
estimated['true'].append(mi_true)
for estimator_name, estimator_value in estimated_test.items():
estimated[estimator_name].append(estimator_value)
value_true = estimated.pop('true')
plt.figure()
plt.plot(parameters, value_true, label='true', ls='--', lw=2, marker='x')
for estimator_name, estimator_value in estimated.items():
plt.plot(parameters, estimator_value, label=estimator_name)
plt.xlabel(xlabel)
plt.ylabel('Estimated Mutual Information, bits')
title = f"{method_name}: len(X)={n_samples}"
if n_features is not None:
title = f"{title}, dim(X)={n_features}"
plt.title(title)
plt.legend()
plt.savefig(IMAGES_MUTUAL_INFO_CLASSIFIER_DIR / f"{method_name}.png")
# plt.show()
if __name__ == '__main__':
set_seed(26)
# plot_gmm()
mi_test(method='sample_gaussian_mixture')
mi_test(method='sample_gaussian_mixture_softmax')
mi_test(sample_softmax_argmax)
Timer.checkpoint()
parser = argparse.ArgumentParser(description="ECG classifier")
parser.add_argument("-d",
"--dir",
default="data",
help="Directory with the data")
parser.add_argument("-m",
"--mode",
help="Script working mode. One of [train, classify]")
parser.add_argument("-r",
"--record",
default="",
help="Name of the record to be classified")
parser.add_argument("-o",
"--output",
default="answers.txt",
help="File where to write classification results")
args = parser.parse_args()
set_seed(42)
if args.mode == "train":
logger.enable_logging('ecg', True)
train(args)
elif args.mode == "classify":
if len(args.record) > 0:
main_classify_single(args)
else:
main_classify_all(args)
else:
parser.print_help()