DOsinga
commented
5 years ago Hi,
You should be able to get this done similarly to how the inverse image search works using Postgres or you can use Scikit-Learn's k-nearest neighbors if you want to do this in memory.
Hope this helps!
Hello,
In example 06.1 (Question matching), the sim_model calculates the similarity between two embeddings - those of the title and those of the body. If I'd like to predict the closest matching body given a title, is there a better way to do this other than loop over every possible combination of [title, body], extract the corresponding similarity, then output the body corresponding to the highest similarity? For large datasets such an approach would be far too slow.