Change paradigm of detection methods
CTAP offers two kinds of noise detection functionality:
- statistical functions, like Hurst exponent or variance, which are generally taken from prior work (e.g. ADJUST, FASTER) and made available using methods of the prior work, sometimes bundled together (e.g. FASTER offers 5 independent component functions in one method)
- 'enhancements' to above, in two basic forms: a. multivariate outlier detection: currently works only with FASTER functions, combining them using Mahalanobis distance b. iteration or recursion with testing: this allows functions to be applied more than once, and the outcome tested to determine a stopping condition
The proposal is to (a) rewrite the CTAP_detect_bad_* methods to allow all functions from 1. to be accessed independently; and then (b) allow any set of functions to be combined using 2a. (multivariate outlier), and/or 2b. (iteration).
2b. would be possibly overlapping with the parameter sweep idea of CTAP-HYDRA.