Automunge

- corrected one of the support functions from 4.83
- to ensure that if we are following chains of functionpointers
- the defaultparams entries from intermediate links are incorporated into the final result
- also added a few code comments for clarity

- improved transformation category specification in user passed processdict data structure
- incorporated 'functionpointer' option
- so user can specify the set of transfomation functions associated with a transformation category without having to dig into code-base to identify naming conventions for existing transformation functions
- now processdict entry can instead just identify a transformation category in functionpointer whose transformation functions the new category would like to match
- such as to automaticaly populate with entries for dualprocess, singleprocess, postprocess, inverseprocess, and info_retention
- defaultparam entries are also accessed, and if the new category specification contains any redundant defaultparam entries with the pointer category the new category entries will take precedence
- as an example, if we previously wanted to define a processdict entry for a new transformation category that reused transformation functions of the mnmx category, it would have looked something like this:
{'newt' : {'dualprocess' : am.process_mnmx_class, \
'singleprocess' : None, \
'postprocess' : am.postprocess_mnmx_class, \
'inverseprocess' : am.inverseprocess_mnmx, \
'info_retention' : True, \
'NArowtype' : 'positivenumeric', \
'MLinfilltype' : 'numeric', \
'labelctgy' : 'mnmx'}}
- in the new convention with functionpointer entry, we can now more easily consistently specify as:
{'newt' : {'functionpointer':'mnmx', \
'NArowtype' : 'positivenumeric', \
'MLinfilltype' : 'numeric', \
'labelctgy' : 'mnmx'}}
- also perormed a few misc code comment cleanups

- another accomodation for small data sets, this time with the eval_ratio parameter
- eval_ratio is meant to improve sampling efficiency for evaluating data properties
- by only inspecting subset of data
- new convention is that eval_ratio is only applied when a training set has > 2,000 rows

- rethinking the PCA heuristic
- which was in place to apply dimensionality reductions automatically when features > 0.5 number of rows
- I was finding that this heuristic was kind of cumbersome when working with small data sets to run experiments
- in which case the heuristic had to be turned off with a ML_cmnd
- and decided from a usability standpoint would be better to make this heuristic optional instead of default
- so new default value for PCAn_components is False to completely turn off PCA
- if you want to run the heuristic and only apply PCA for that scenario, can pass PCAn_components = None

- rewrite of the dxd2 transform
- a review found that I had kind of bungled the implementation
- when had performed the conversion to accepting periods parameter
- current form is consistent with original intent

- reintroduced a varient of the standard deviation bins that assume input data is already normalized
- by incorporating boolean parameter normalizedinput into bins and bsor
- such as may reduce the computational overhead associated with the binstransform parameter