Base trait for all Evaluation functions.
This class implements the slice sampling algorithm.
This class implements the slice sampling algorithm. Slice sampling is an Markov chain Monte Carlo algorithm for sampling from an arbitrary continuous distribution function.
Consider the ASCII-normal distribution function shown below.
/-\ /-\ f(x) /-\ /-\ /-\ / \ / |\ / |\ / \ / \ / \ / | \ /---|-\ /-x'--\ / \ --/ x \-- --/ x \-- --/ x \-- --/ \-- --/ x' \--
1 2 3 4 5
The procedure for drawing a sample from the function is as follows:
1) Start with an initial point x in the domain (e.g. the sample from a previous iteration). 2) Sample a vertical value uniformly between zero and f(x). 3) Step out horizontally in both directions from the value obtained in step 2 until reaching the boundary of the function. This is the "slice". 4) Sample uniformly from the slice to obtain the new point x'. 5) Back to step 1 with the new point.
Base trait for all Evaluation functions.
An evaluation function is the integration point between the hyper-parameter tuning module and an estimator, or any system that can unpack a vector of values and produce a real evaluation.