Closed ZhikangLai closed 2 months ago
additionally, I want to know "lightning_logs" folder is useful? it took up quite a lot of space in my computer
Hey. We use pytorch lightning's Trainer which sets those defaults. You can get rid of the progress bar by passing enable_progress_bar=False
through the config. The lightning_logs directory also comes from that and it saves the losses during training (which you can then visualize through tensorboard) and you can get rid of those by passing logger=False
through the config.
So by changing the code to the following you shouldn't see those anymore:
rnn_config = {
**AutoRNN.get_default_config(h=h, backend='ray'),
'enable_progress_bar': False,
'logger': False,
}
models = [AutoRNN(h=h, loss=MSE(), config=rnn_config, num_samples=10)]
Hey. We use pytorch lightning's Trainer which sets those defaults. You can get rid of the progress bar by passing
enable_progress_bar=False
through the config. The lightning_logs directory also comes from that and it saves the losses during training (which you can then visualize through tensorboard) and you can get rid of those by passinglogger=False
through the config.So by changing the code to the following you shouldn't see those anymore:
rnn_config = { **AutoRNN.get_default_config(h=h, backend='ray'), 'enable_progress_bar': False, 'logger': False, } models = [AutoRNN(h=h, loss=MSE(), config=rnn_config, num_samples=10)]
Thank you very much for your response. That's working.
Description
Hello there. Neuralforecast is excellent package. But I think some functions need to be improved. For instance, When I call the AutoModel,such as AutoRNN, and I will out lots of training message(train bar). Not everyone needs this information. So I think the creator of this toolbox can add this feature which can let user choose for themselves whether to ignore information about the training process.
Use case
h = 1 models = [AutoRNN(h = h, loss=MSE(), num_samples=10)] auto_nf = NeuralForecast(models=models, freq='D',local_scaler_type = 'minmax') auto_cv_df = auto_nf.cross_validation(df=No4, refit=0, n_windows = len(test_data))
and it will outbreak lots of unnecessary information for me:![1712739164275](https://github.com/Nixtla/neuralforecast/assets/110596898/67e3bd6a-4c3f-442a-8065-f3563b1f99bd)