For each model a figure showing the best result for each setting

pykeen / benchmarking

📊 Results from the reproducibility and benchmarking studies presented in "Bringing Light Into the Dark: A Large-scale Evaluation of Knowledge Graph Embedding Models Under a Unified Framework" (http://arxiv.org/abs/2006.13365)

MIT License

35 stars 4 forks source link

For each model a figure showing the best result for each setting #6

Closed mali-git closed 4 years ago

mali-git commented 4 years ago

For each model we would need a figure that shows the best results for each setting, e.g. TransE using OWA + BCE + Inverse Triples, LCWA + BCE + Inverse Triples ....

cthoyt commented 4 years ago

Done in 1b6c5933, see https://github.com/mali-git/pykeen_experimental_results/tree/master/ablation/summary/modelsummary. They don't look very good because they have tight distributions compared to how spread out they are. Probably would be good to sort them as well, but this isn't really conveying anything meaningful in my opinion since there are combanatorially many things to compare on the y axes