Access to stored vector about hnswlib

[yurymalkov]

Hi @rafael-aero,

You can access the data by using the method char *getDataByInternalId(tableint internal_id) which returns the pointer to the first element in the vector. To get an internal ID of an element you should store the result of HNSW's addPoint(void *data_point, labeltype label, int level).

Right now the element's internal IDs never change. But this probably might be changed in the future.

[theolivenbaum]

awesome, thanks for the super quick reply 👍
I'll keep in mind for future versions to check if this is still the case.

[glennkroegel]

Is there a way of doing this with the python bindings?

The method nmslib.getDataPoint(index, i) doesn't seem to work if you are loading a previously saved index. Executing nmslib.getDataPointQty(index) returns 0 even if the data is loaded (as index.knnQuery does return results).

[yurymalkov]

@glennkroegel Shouldn't be so. I'll look into it.

[yurymalkov]

@glennkroegel It seems that you are referencing nmslib, not hnswlib. hnswlib has a different function with similar functionality (get_items). It also should work after saving/loading the index without any problems.

In nmslib's hnsw you need to reload all elements separately from loading the index itself in order to access them (hope @searchivarius will correct me if I am wrong). This is because hnsw in nmslib is using its own data structures.