This is a repo for MIDX sampler applied in other domains like NLP, Knowledge Graph, Sequential Recommendation and Extreme Classification.
python run.py -m=LSTM -d=penntreebank --sampler=midx-uni
Task | Model | Dataset |
---|---|---|
LanguageModeling | [LSTM, Transformer] | [penntreebank, wikitext-2, wikitext-103] |
KnowledgeGraph | [HouseE, TransE] | [fb15k, wn18, yago3-10] |
ExtremeClassification | [EcFc, ] | [AmazonCat-13k, Delicious-200k, WikiLSHTC] |
Optional Sampler: [None, 'midx-uni', 'midx-pop'] (Default: None)
when sampler is None, FullSoftmax is used as loss function, otherwise sampled softmax is used.