Learning Sum-Product Networks with Direct and Indirect Variable Interactions
Authors: Amirmohammad Rooshenas, Daniel Lowd
ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In experiments on 20 benchmark datasets, we find that the combination of direct and indirect interactions leads to significantly better accuracy than several state-of-the-art algorithms for learning SPNs and other tractable models. |
| Researcher Affiliation | Academia | Amirmohammad Rooshenas PEDRAM@CS.UOREGON.EDU Daniel Lowd LOWD@CS.UOREGON.EDU Department of Computer and Information Science, University of Oregon |
| Pseudocode | Yes | Algorithm 1 Algorithm for learning a SPAC. function ID-SPN(T) input: Set of training examples, T; set of variables V output: Learned SPAC model n Learn AC(T, V ). SPAC n N leaves of SPAC // which includes only n while N = do... |
| Open Source Code | No | The paper does not state that its own source code (for ID-SPN) is open-source or provide a link to it. It only refers to a third-party code for LTM: 'http://people.csail.mit.edu/myungjin/latent Tree.html'. |
| Open Datasets | Yes | We evaluated ID-SPN on 20 datasets illustrated in Table 1.a with 16 to 1556 binary-valued variables. These datasets or a subset of them also have been used in (Davis & Domingos, 2010; Lowd & Davis, 2010; Haaren & Davis, 2012; Lowd & Rooshenas, 2013; Gens & Domingos, 2013). |
| Dataset Splits | Yes | Table 1.a shows: 'Dataset Var# Train Valid Test NLTCS 16 16181 2157 3236 MSNBC 17 291326 38843 58265 KDDCup 2000 64 180092 19907 34955 Plants 69 17412 2321 3482 Audio 100 15000 2000 3000 Jester 100 9000 1000 4116 Netflix 100 15000 2000 3000 Accidents 111 12758 1700 2551 Retail 135 22041 2938 4408 Pumsb-star 163 12262 1635 2452 DNA 180 1600 400 1186 Kosarek 190 33375 4450 6675 MSWeb 294 29441 32750 5000 Book 500 8700 1159 1739 Each Movie 500 4524 1002 591 Web KB 839 2803 558 838 Reuters-52 889 6532 1028 1540 20 Newsgroup 910 11293 3764 3764 BBC 1058 1670 225 330 Ad 1556 2461 327 491'. |
| Hardware Specification | Yes | We bounded the learning time of all methods to 24 hours, and we ran our experiments, including learning, tuning, and testing, on an Intel(R) Xeon(R) CPU X5650@2.67GHz. |
| Software Dependencies | No | The paper mentions several tools and algorithms (e.g., 'Win Mine toolkit', 'Libra toolkit', 'ACMN algorithm') but does not specify software versions for any dependencies like programming languages or libraries. |
| Experiment Setup | Yes | For ACMN, we used an L1 prior of 0.1, 1, and 5, and a Gaussian prior with a standard deviation of 0.1, 0.5, and 1. We also used a split penalty of 2, 5, 10 and maximum edge number of 2 million. For ID-SPN, we selected C from 1.0, 2.0, 5.0, 8.0, 10.0, 15.0 and 20.0, and SP from 5, 8, 10, 15, 20, 25, and 30. We selected the pair setting of (Cmin, SPmin) between (0.01, 1) and (1, 2), and ME and ME min are 2M and 200k edges, respectively. We used similar Gaussian priors with a standard deviation of 0.1, 0.3, 0.5, 0.8, 1.0, or 2.0 for learning all AC nodes. For learning sum nodes, the cluster penalty λ is selected from 0.1, 0.2, 0.4, 0.6 and 0.8, the number of clusters from 5, 10, and 20, and we restarted EM 5 times. When there were fewer than 50 samples, we did not learn additional sum nodes, and when there were fewer than 10 variables, we did not learn additional product nodes. Finally, we limited the number of main iterations of IDSPN to 5, 10, or 15. |