models
models copied to clipboard
Make ranking-metric calculation more efficient in item-retrieval
Currently we calculate the metrics for both the items in the dataset as well as the sampled negatives. We shouldn’t calculate it for the sampled negatives.