Automunge

- new Q Notation family of transforms available as qbt1 / qbt2 / qbt3 / qbt4
- where encoding is to binary with seperate registers for integers, fractionals, and sign
- transforms accept parameters suffix / integer_bits / fractional_bits / sign_bit
- parameters designate the qubit capacity of each register,
- defaulting for qbt1 to {'sign_bit' : True, 'integer_bits' : 3, 'fractional_bits' : 12}
- (the defaults are arbitrary representing a compact register for range +/- 8.0000)
- qbt2 is for signed integers defaulting to {'sign_bit' : True, 'integer_bits' : 15, 'fractional_bits' : 0}
- qbt3 is for unsigned floats defaulting to {'sign_bit' : False, 'integer_bits' : 3, 'fractional_bits' : 12}
- qbt4 is for unsigned integers defaulting to {'sign_bit' : False, 'integer_bits' : 15, 'fractional_bits' : 0}
- and with suffix corresponding to category key for each
- the expectation is in many workflows users may wish to deviate from default register counts, these are just starting points
- register sizes were selected to accomodate z-score normalized data with +/-6 standard deviations from mean and approx 4 sig figures in decimals
- requiring 16 qubits in base qbt1 configuration for signed floats
- missing data and overflows default to zero infill
- if markers are needed for missing data can turn on NArw_marker parameter
- (NArw won't pick up overflow cases, so care should be taken for adequate register size)
- for example, with default parameters an input column 'floats' will return columns:
- ['floats_qbt1_sign', 'floats_qbt1_2^2', 'floats_qbt1_2^1', 'floats_qbt1_2^0', 'floats_qbt1_2^-1', 'floats_qbt1_2^-2', 'floats_qbt1_2^-3', 'floats_qbt1_2^-4', 'floats_qbt1_2^-5', 'floats_qbt1_2^-6', 'floats_qbt1_2^-7', 'floats_qbt1_2^-8', 'floats_qbt1_2^-9', 'floats_qbt1_2^-10', 'floats_qbt1_2^-11', 'floats_qbt1_2^-12']
- inversion also supported
- excluded from ML infill for now
- Q notation was inspired by discussions in "Programming Quantum Computers: Essential Algorithms and Code Samples" by Eric R. Johnston, Nic Harrigan, and Mercedes Gimeno-Segovia

- revisiting defaults for cv scoring in grid search hyperparameter tuning
- for ML infill hyperparameter tuning in cases of classification
- reverting performance metric from f1 score back to accuracy
- (the problem with f1 is if folds split doesn't have fully represented activations triggers printouts)
- iterate, iterate, and then iterate some more

- a simplification of the index column populated in ID sets
- now index column header string defaults to 'Automunge_index'
- with exception of reverting to the previous convention 'Automunge_index_'
- (where is the 12 digit random integer associated with the application)
- for remote edge case with a column with header 'Automunge_index'
- is already found in carved out ID columns
- this should make the ID sets a little more user friendly
- with a index header now known in advance

- found and fixed a bug introduced in 5.61 associated with feature selection
- just needed to move the randomseed initialization a few lines up
- also found a new edge case for parsenumeric NArowtype
- fixed by converting entries to strings for parsing operation
- and in the process realized that could consolidate three of NArowtype categories into a single type
- so now NArowtypes 'parsenumeric', 'parsenumeric_commas', and 'parsenumeric_EU'
- are all grouped together as NArowtype 'parsenumeric'
- which registers NArw activations when numeric characters aren't present in an entry

- found and fixed bug associated with trainID_column parameter
- associated with cases when passed as a list of columns
- used as an opportunity for some cleanups to ID / index column processing
- in both automunge and postmunge
- including some code comments
- and reductions of redundancy
- ID code portion is a little more legible now I think

- another iterative improvement to transformdict specification
- now primitives with single entry can have entry passed as a string
- instead of string embedded in list
- (in other words can omit list brackets for primitives with single entry)
- this is just a little more intuitive, can already do this in assigncat and assigninfill
- also another validation added for data types of primitive entries