Why optimize summed log densities rather than multiplying thousands of small densities?
StatisticsLevel: IntermediateLikelihood and fitted distributions
Why optimize summed log densities rather than multiplying thousands of small densities?
Hints
- Log is monotone.
- Products shrink rapidly.
Answer
Products can underflow. Logarithms turn independent likelihood products into sums while preserving the maximizing parameter values.
Source and licence: SciPy — distribution fitting and likelihood · BSD-3-Clause · answer: Quantoros editorial