Differentiable Structure Learning with Partial Orders
Authors: Taiyu Ban, Lyuzhou Chen, Xiangyu Wang, Xin Wang, Derui Lyu, Huanhuan Chen
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This result, together with a comprehensive evaluation on synthetic cases, demonstrates our method s ability to effectively improve differentiable structure learning with partial orders. This result, together with a comprehensive evaluation on synthetic cases, demonstrates our method s ability to effectively improve differentiable structure learning with partial orders. |
| Researcher Affiliation | Academia | University of Science and Technology of China {banty, clz31415, wz520, drlv}@mail.ustc.edu.cn {sa312, hchen}@ustc.edu.cn |
| Pseudocode | Yes | Appendix D Implementation of the Proposed Method This section presents the psudocodes of the implementations of the proposed method. Algorithm 1 Partial Order Constraint-based Differentiable Structure Learning... Algorithm 2 Augmented Acyclicity Characterization Function... Algorithm 3 Find All Maximal Paths in a DAG... Algorithm 4 Derive Transitive Reduction of a Relation... |
| Open Source Code | Yes | Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: All the codes of implementations and evaluations are provided in the supplementary file. |
| Open Datasets | Yes | The dataset provided by Sachs et al. [2005] consists of continuous measurements of protein and phospholipid expression levels in human immune system cells. It is frequently used as a benchmark in graphical models... and Random DAGs are generated using Erdös-Rényi (ER) and scale-free (SF) models with node degrees in {2, 4} and numbers of nodes d in {20, 30, 50}. |
| Dataset Splits | No | The paper describes generating synthetic datasets and using subsets of a real-world dataset for testing ('we selected the first s data samples for testing'), but it does not specify explicit training, validation, and test splits with percentages or sample counts for all experiments. For synthetic data, samples are generated, not split from an existing dataset. |
| Hardware Specification | Yes | For computational resources, linear NOTEARS and DAGMA are executed on a 32-core AMD Ryzen 9 7950X CPU at 4.5GHz, while NOTEARS-MLP uses an NVIDIA Ge Force RTX 3090 GPU, both with a 32GB memory limit. |
| Software Dependencies | No | The paper mentions software like NOTEARS, NOTEARS-MLP, and DAGMA, but it does not specify version numbers for these or any other ancillary software components or libraries (e.g., Python, PyTorch, TensorFlow) used in the experiments. |
| Experiment Setup | Yes | The hyper-parameters of algorithms include: the threshold γ = 0.3 for the existence of an edge, the weight λ1 = 0.03 for L1 regularization term, the weight λ2 = 0.01 for L2 regularization term, the weight τ = 1 in the proposed augmented acyclicity term, the maximum ρ value ρmax = 1016 for augment Lagrangian method. The NOTEARS-MLP uses an MLP with an input dimension d, a hidden layer of size d 10, and an output layer of size d, where d is the number of variables. |