Automunge

- performed a comprehensive audit of edge case printouts
- to ensure validation test results are being recorded in appropriate reports
- identified a few cases where validation checks were missing recorded results
- so added new entries to postprocess_dict['miscparameters_results']
- as trainID_column_valresult, testID_column_valresult, evalcat_valresult, validate_traintest_columnnumbercompare, validate_traintest_columnlabelscompare, validate_redundantcolumnlabels, validate_traintest_columnorder, numbercategoryheuristic_valresult
- and added new entries to postreports_dict['pm_miscparameters_results']
- as testID_column_valresult, validate_traintest_columnlabelscompare, validate_traintest_columnorder, validate_labelscolumn_string
- in the process developed some documentation defining various validation checks
- which am keeping as internal for time being

- updated postmunge featureeval methods to include support for cases where automunge privacyencode was elected
- updated scikit random forest initializer to include support for parameters ccp_alpha and max_samples (which scikit introduced in 0.22)
- update tweak to pwor transform to circumvent some kind of strange interaction between np.nan serving as key to python dictionary

- a refinement of the conditional data type conversions for ordinal encodings
- which again convert transformation outputs to one of uint8/uint16/uint32
- based on the size of the encoding space
- found that our logic test for selection was off by a few integers
- now will maximally utilize capacity of data types

- new set of validations performed for scenarios with model training
- including ML infill, feature selection, and PCA
- now prior to training data is validated to confirm all valid numeric entries
- which is automatic for most transforms in library, but there are a select few that may return non-numeric or NaN values
- such as excl, copy, shfl, strg
- if validation doesn't pass results in printout message
- validation results are returned with other tests in postprocess_dict['miscparameters_results'] for automunge(.)
- and postreports_dict['pm_miscparameters_results'] for postmunge(.)
- removed some automunge(.) initializations of variables that were no longer used (multicolumntransform_dict and LSfitparams_dict)
- a few code comments added, a few tweaks to printouts
- added inplace support for copy transform

- updated returned data type convention for exc5 and exc8 which are for integer passthrough
- exc5 has mlinfilltype singlct as intended for encoded categoric sets
- exc8 has mlinfilltype integer as intended for continuous integer sets
- previously these were being returned as float data types based on floatprecision designation
- recast to set data type in the transformation, making use of new assignparam parameter integertype
- where integertype defaults to 'singlect' and can also be passed as 'integer'
- where for 'singlct' returned data type is conditional uint based on max in train set which is used as proxy for encoding space, which is the default for exc5
- and for 'integer' returned data type is int32, which is the default for exc8
- updated the logic test for powertransform = 'infill' to distinguish between assigning exc5 and exc8
- previously exc5 was default for integer sets unless for train len unique set > 75% number of rows then cast as exc8
- now new scenario added to exc8 for cases where any negative integers found in train set allowing us to infer set is a continuous integer
- new convention for floatprecision conversion to now only be applied to columns returned from transforms with MLinfilltype in {'numeric', 'concurrent_nmbr'} or columns returned from PCA
- instead of previous convention's use of a type check for float
- which in practice means that now excl passthrough columns will retain their data type from received data instead of defering to floatprecision
- moved the application of floatprecision conversion in automunge and postmunge workflow to take place prior to excl suffix extraction to accomodate the MLinfilltype check
- which as side benefit means don't have to worry about privacy encodings
- finally, an update to methods for populating the column_map report
- which now will include empty set entries for source columns that had all of their derivations consolidated in a PCA or Binary dimensionality reduction

- struck an unneccesary numpy conversion in PCA application
- revisited a heuristic in place for PCA application associated with identifying cases when didn't have enough samples for the number of features (aka the col_row_ratio)
- since this was previously struck in documentation went ahead and initialized variable so it isn't inspected in ML_cmnd
- but left the code in place in case later decide to reintroduce
- improved a convention used in _postprocess_DPod from accessing an upstream normalization_dict to accessing own (just needed to add an additional entry to DPod normalization_dict)
- conducted audit of returned data types from entire library
- where convention is returned continuous sets have data type not addressed in transformation function but are converted based on the floatprecision parameter externally
- boolean integer sets are cast as np.int8
- and ordinal encoded integer sets are given a conditional data type based on size of encoding space as either uint8, uint16, or uint32
- found a few transforms needing update to dtype convention
- recast bsor from int8 to conditional (since bin count is customizable by parameter)
- added missing data type casting for DPbn (int8) and DPod (conditional)
- added some documentation to the read me associated with returned data types in section on automunge(.) returned sets and postmunge(.) returned sets
- then just a few opportunities to consolidate transformations to reduce lines of code:
- consolidated bnry and bnr2 to use of a single comnmon transformation function by adding parameter for infillconvention
- consolidated MADn and MAD3 to use of a single common transformation function by adding parameter for center
- consolidated transformation functions for shft, shf2, and shf3 into a single common trasnformation function (taking advantage of what is now standard accross library allowing redundant transformations to common input column applied with different parameters)