For real neighbor computation the query vector needed to be transposed to compute the dot product (was producing error). Also unit vector function was applied to get valid distances. Lastly, results are now printed with corresponding data IDs so it's easier to compare the approximations with the real values. We get close results when POINTS variable is low (e.g. 100) and as expected things get more approximate with more points.
Still not sure about RandomBinaryProjections because it doesn't seem to generate any candidates - or is that by design?
For real neighbor computation the query vector needed to be transposed to compute the dot product (was producing error). Also unit vector function was applied to get valid distances. Lastly, results are now printed with corresponding data IDs so it's easier to compare the approximations with the real values. We get close results when POINTS variable is low (e.g. 100) and as expected things get more approximate with more points.
Still not sure about RandomBinaryProjections because it doesn't seem to generate any candidates - or is that by design?