|
I'm using an existing implementation of LSH and I'm not quite sure about how to determine the right number of hash tables and key size. Is there a relation between the size of data that is being hashed to the table and key sizes? |
|
I assume you're doing LSH for k-NN. Is that true? I give a detailed description here of how to select the hyperparameters for LSH. Please let me know if you have more questions. In fact, I'm surprised that no one has built an LSH implementation that helps you with hyperparameter selection. If you build a script on top of your existing implementation, that measures speed and accuracy tradeoffs for the different choices of hyperparams, I bet that a lot of people would use it. |