Feature Programming for Multivariate Time Series Prediction

Authors: Alex Daniel Reneau, Jerry Yao-Chieh Hu, Ammar Gilani, Han Liu

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerically, we validate the efficacy of our method on several synthetic and real-world noisy time series datasets.
Researcher Affiliation Academia 1Department of Computer Science, University of Northwestern, Evanston, USA 2Department of Statistics and Data Science, University of Northwestern, Evanston, USA.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at Git Hub; the most updated version is available on ar Xiv with a full list of authors, including Chenwei Xu and Weijian Li. We kindly request that citations refer to the ar Xiv version: https://arxiv.org/abs/2306.06252.
Open Datasets Yes The data utilized in our experiments consists of a synthetic dataset constructed to adhere to the assumptions of our method, as well as an electricity dataset, a traffic dataset, and a taxi dataset. ... Taxi Dataset: We use the TLC Trip Record Dataset... Electricity Dataset: We use the UCI Electricity Load Diagrams Dataset... Traffic Dataset: We use the UCI PEM-SF Traffic Dataset...
Dataset Splits No Each of these datasets is partitioned in an 80/20 ratio to derive our training data (known as insample data) and testing data (referred to as out-of-sample data).
Hardware Specification Yes Platforms: The GPUs and CPUs used to conduct experiments are NVIDIA GEFORCE RTX 2080 Ti and INTEL XEON SILVER 4214 @ 2.20GHz.
Software Dependencies No The paper mentions software like 'DART' and 'XGBoost', 'Light GBM', 'Transformer', 'TFT', 'TCN', 'N-BEATS' but does not specify their version numbers.
Experiment Setup Yes Hyperparameter optimization is conducted via random search for 100 iterations. learning rate: 0.01, 0.001, 0.0001, 0.00001 batch size: 64, 128, 256, 512, feature dim hidden size: 64, 128, 512, 1024, 2048 num epochs: we use early stopping. ... We use an Adam optimizer with learning rate lr = 10 5 for training. The coefficients of Adam optimizer, betas, are set to (0.9, 0.999).