Conformal Time-series Forecasting
Authors: Kamile Stankeviciute, Ahmed M. Alaa, Mihaela van der Schaar
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we showcase the performance of the conformal forecasting RNN (CF-RNN) model against three baselines: the frequentist blockwise jackknife RNN (BJ-RNN) [21], the multi-quantile RNN (MQ-RNN) [3], and the Monte Carlo dropout-based RNN (DP-RNN) [26]. ... We first present the performance of CF-RNNs on synthetic data with controlled properties. ... Finally, we compare the performance of CF-RNNs with the remaining two baselines on three real-world medical datasets. |
| Researcher Affiliation | Academia | Kamil e Stankeviˇci ut e University of Oxford University of Cambridge ks830@cam.ac.uk Ahmed M. Alaa University of California, Los Angeles ahmedmalaa@ucla.edu Mihaela van der Schaar University of Cambridge University of California, Los Angeles The Alan Turing Institute mv472@cam.ac.uk |
| Pseudocode | Yes | Algorithm 1 Conformal forecasting RNN (CF-RNN) |
| Open Source Code | Yes | The code is available at github.com/kamilest/conformal-rnn. |
| Open Datasets | Yes | MIMIC-III [52] dataset, electroencephalography (EEG) dataset from the UCI machine learning repository [53], daily COVID-19 cases within the United Kingdom local authority districts [54]. All datasets are publicly available and the medical data is anonymised. |
| Dataset Splits | Yes | We train the models on 2000 training sequences (with CF-RNNs splitting this dataset into 1000 true training and 1000 calibration sequences) and The inductive variant of CP operates by splitting the training set into the proper training set of size n and a calibration set of size m: D = Dtrain Dcal. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments. |
| Software Dependencies | No | The paper mentions using LSTM but does not provide specific version numbers for any software dependencies or libraries used in the experiments. |
| Experiment Setup | Yes | The RNN hyperparameters for the networks underlying the uncertainty estimation models are fixed in order to ensure fair comparison, and largely follow those provided in previous work [21]. These are detailed in the Appendix B along with the time-series model parameters. |