Matthias Petri

Not sure if this is the best way to report this but currently when I try to use the deltalake crate with datafusion: ``` datafusion = "7.0.0" deltalake = {...

you are right. I'm not really maintaining this anymore. You should probably use the sdsl-lite library (https://github.com/simongog/sdsl-lite) which contains many more succinct text indexes besides the FM-index and uses proper...

From my understanding the C++ code does not contain parameter optimization. Only the python code called [here](https://github.com/dblalock/bolt/blob/master/experiments/python/vquantizers.py#L606=) does. I have been trying to piece together what pieces from the python...

> Since there are 16 rows, but we only get 15 values per codebook from python, I have zeroed out the last row at all columns I have the same...

@fjrdev do you have some code to share that extracts the python values and imports into the c++ codebase?

@fjrdev I ported the python code to rust that produces equivalent results. However, I'm now trying to incorporate the more efficient c++ code and I'm encountering more and more issues....

Maybe I don't really understand the algorithm correctly but from my understand you use the split vals to walk down a binary tree updating the codes with 2 * code...

Actually looking at the code some more I see we have a for loop over blocks and inside that a for loop over split vals so that would mean we...

@dumpinfo have you managed to port the python learning code to c++ or at least were able to extract the learnt parameters into the c++ codebase?

To put some context into this: I was creating a collection and processing the documents (where it outputs [Batch 1234] Processed documents) took the same time as the "remapping IDs"...