Trafformer: Unify Time and Space in Traffic Prediction
Authors: Di Jin, Jiayi Shi, Rui Wang, Yawen Li, Yuxiao Huang, Yu-Bin Yang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on two traffic datasets demonstrate that Trafformer outperforms existing methods and provides a promising future direction for the spatial-temporal traffic prediction problem. |
| Researcher Affiliation | Academia | 1College of Intelligence and Computing, Tianjin University, Tianjin, P.R. China 2State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, P.R. China 3School of Economics and Management, Beijing University of Posts and Telecommunications, Beijing, P.R. China 4Columbian College of Arts Sciences, George Washington University, Washington, D.C., USA |
| Pseudocode | No | The paper describes the model architecture and strategies in text but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement or link indicating the public availability of its source code. |
| Open Datasets | Yes | We evaluate the performance of our proposed model using two real-world datasets, the public transportation network datasets METR-LA and PEMS-Bay. |
| Dataset Splits | Yes | During the experiment, both datasets are sorted in ascending chronological order and are split into three parts for training, validation, and testing, which account for 70%, 10%, and 20% respectively. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper mentions using 'Adam optimizer' but does not specify any software names with version numbers for replication (e.g., Python, PyTorch, TensorFlow versions or specific library versions). |
| Experiment Setup | Yes | For Transformer and Trafformer, we set the encoder layer to 2 and decoder layer to 1. The hidden dimension for each attention layer is set to 32. For Trafformer, the batch size for METR-LA is set to 11 and the batch size for PEMS-Bay is set to 5. All learnable parameters are initialized with a normal distribution. We use Adam optimizer with an initial learing rate of 0.002. We test 3 settings for dropout rate (0, 0.05, 0.2) and set the rate to 0. |