Open vpodpecan opened 3 years ago
Same with ImageSpace. The example (https://github.com/facebookresearch/StarSpace/#imagespace-learning-image-and-label-embeddings) reports an accuracy of ~96%, but when I run the example script, I merely get hits@1 of 0.4896. Strange..
When running the
classification_ag_news.sh
example I get hit@1=0.46. However, the SparSpace paper reports 91.5% accuracy. What could be the reason for this big difference?