filippomb / ensemble-conformalized-quantile-regression

Valid and adaptive prediction intervals for probabilistic time series forecasting

Python implementation of the ensemble conformalized quantile regression (EnCQR) algorithm, as presented in the original paper. EnCQR allows to generate accurate prediction intervals when predicting a time series with a generic regression algorithm for time series forecasting, such as a Recurrent Neural Network or Random Forest.

---

Example of usage

The code in main_EnCQR.py shows a quick example of how to perform probabilistic forecasting with EnCQR.

A detailed tutorial can be found in this notebook, which explaines how the dataset are preprocessed and shows the differences between different regression models (LSTM, Temporal Convolutional Network, and Random Forest), which can be used as base models in the EnCQR ensemble.

---

Citation

Consider citing the original paper if you are using EnCQR in your reasearch

@misc{jensen2022ensemble,
      title={Ensemble Conformalized Quantile Regression for Probabilistic Time Series Forecasting}, 
      author={Vilde Jensen and Filippo Maria Bianchi and Stian Norman Anfinsen},
      year={2022},
      eprint={2202.08756},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

ensemble-conformalized-quantile-regression's Issues

In the preprocessing.py, when creating subsets, it seems there's no operation about the discarded sequence?

About the outcome of the prediction

Hello, thank you for sharing the code！
However, in the process of learning, I had the following problems: After training the network with the verification set, I predicted the test set, and found that the test set predicted the same result every day. What is the cause of this problem?

Looking forward to your reply！

Predict future results

Hi, I'm sorry to bother you again.
I have a question to ask you, I wonder if this method can predict future values.For example, can I predict the future values for some time in the future, of course, I know I can't verify that future values，I just want to know if it can be implemented.Looking forward to your reply！！Thanks a lot！！

集成一致性分位数回归参数设置问题

我想问一下集成一致性分位数回归代码里，time_steps_in=24,time_steps_out=6这里。如果数据是每分钟一条，这里该如何表示呢。

Does it support predicting multidimensional y?

Interesting work. In your example, the prediction data is only 1-dimensional, and I wonder if the model supports multidimensional y. thank you

Idiscard sequence