DyRep: Learning Representations over Dynamic Graphs
Authors: Rakshit Trivedi, Mehrdad Farajtabar, Prasenjeet Biswal, Hongyuan Zha
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that Dy Rep outperforms state-of-the-art baselines for dynamic link prediction and time prediction tasks and present extensive qualitative insights into our framework. 5 EXPERIMENTS |
| Researcher Affiliation | Collaboration | 1Georgia Institute of Technology 2Deep Mind |
| Pseudocode | Yes | Algorithm 1 Update Algorithm for S and A Algorithm 2 Computation of integral term in L for a mini-batch |
| Open Source Code | No | The paper does not include an explicit statement or a link providing concrete access to the source code for the described methodology. |
| Open Datasets | Yes | Social Evolution Dataset released by MIT Human Dynamics Lab #nodes: 83, #Initial Associations: 376, #Final Associations: 791, #Communications: 2016339 and Clustering Coefficient: 0.548. Github Dataset available at Github Archive #nodes: 12328, #Initial Associations: 70640, #Final Associations: 166565, #Communications: 604649 and Clustering Coefficient: 0.087. |
| Dataset Splits | No | The paper describes a train/test split for evaluation ('70/30 (train/test) split' and '65/35 (train/test) split') but does not explicitly mention specific validation dataset splits or percentages for hyperparameter tuning. |
| Hardware Specification | No | The paper does not provide specific hardware details (like exact GPU/CPU models, processor types, or memory amounts) used for running its main experiments. It only vaguely refers to 'Our machine' in Appendix G.2 for t-SNE visualization setup, which is not the experimental hardware for training. |
| Software Dependencies | No | The paper mentions programming languages and frameworks used for baselines (e.g., 'C++', 'Tensorflow', 'Python', 'Keras with Theano backend') but does not specify exact version numbers for these software dependencies or for their own implementation. |
| Experiment Setup | Yes | For social dataset: Num nodes = 100, Num Dynamics = 2, bptt (sequence length) = 200, embed_size = 32, hidden_unit_size = 32, nsamples (for survival) = 5, gradient_clip = 100 and no dropout. For github dataset: Num nodes = 12328, Num Dynamics = 2, bptt (sequence length) = 300, embed_size = 256, hidden_unit_size = 256, nsamples (for survival) = 5, gradient_clip = 100. |