Reducing Execution Time

[JacksonBurns]

Using Python's built-in cProfile, I am going to identify some bottlenecks and hopefully remedy them.

[JacksonBurns]

First fix (above) -- similarity_measure was calling sum on both descriptors to check if they had any active bits. According to the profiler, this was taking up a startling 94% of execution time.

Switching this call to run with np.any is 90% faster and removes this bottleneck.

[JacksonBurns]

Replacing flatten() with ravel() also sees a non-zero speed increase.

[JacksonBurns]

Using hasattr is slower than getattr (for the positive case, at least).

[JacksonBurns]

Remaining substantial time sinks are in basic handling of rdkit and numpy data -- the only way to effectively speed these up is to manipulate the native C objects, which is a task for molSim 2.0.0
Merging, and re-releasing as 1.0.0-rc.3 since this has significant speedup.