Set Functions for Time Series
Authors: Max Horn, Michael Moor, Christian Bock, Bastian Rieck, Karsten Borgwardt
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We extensively compare our method with existing algorithms on multiple healthcare time series datasets and demonstrate that it performs competitively whilst significantly reducing runtime. |
| Researcher Affiliation | Academia | 1Department of Biosystems Science and Engineering, ETH Zurich, 4058 Basel, Switzerland 2SIB Swiss Institute of Bioinformatics, Switzerland. |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We executed all experiments and implementations in a unified and modular code base, which we make available to the community. We provide two dedicated packages (i) for automatic downloading and preprocessing of the datasets according to the splits used in this work and (ii) for training the introduced method and baselines to which we compare in the following. We make both publicly available4. https://github.com/Borgwardt Lab/Set_ Functions_for_Time_Series |
| Open Datasets | Yes | MIMIC-III (Johnson et al., 2016) is a widely-used, freely-accessible dataset... The 2012 Physionet challenge dataset (Goldberger et al., 2000)... Reyna et al. (2020) launched a challenge for the early detection of sepsis from clinical data. |
| Dataset Splits | Yes | Training was stopped after 30 epochs without improvement of the area under the precision recall curve (AUPRC) on the validation data for the mortality prediction tasks... The train, validation, and test splits were the same for all models and all evaluations. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments. It mentions general computing concepts like "runtime" and "GPU implementations" but no concrete specifications. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with versions) needed to replicate the experiment. |
| Experiment Setup | Yes | To mitigate the problem of unbalanced datasets, all models were trained on balanced batches of the training data rather than utilizing class weights... In our experiments we set the number of optimizer steps per epoch to be the minimum of the number of steps required for seeing all samples from the majority class and the number of steps required to see each samples from the minority class three times. Training was stopped after 30 epochs without improvement... The hyperparameters with the best overall validation performance were selected... we executed hyperparameter searches for each model on each dataset, composed of uniformly sampling 20 parameters according to Appendix A.4. |