The gain comes mostly due to:
1) skipping the construction of an additional matrix Q that holds
normalized distribution.
2) being consistent about using TypedArrays (if available)
For example, a single step in clustering 3000 points was reduced from
540ms to 240ms on a MacBook Pro using Chrome.
The gain comes mostly due to: 1) skipping the construction of an additional matrix Q that holds normalized distribution. 2) being consistent about using TypedArrays (if available)
For example, a single step in clustering 3000 points was reduced from 540ms to 240ms on a MacBook Pro using Chrome.