Automunge

- Went for a walk and realized the 3.42 validations were incomplete, needed to move to before the application of circleoflife function in order to include in validation those columns which are subsequently deleted as part of a replacement operation. Complete.

- Added some new validations to identify edge cases where returned columns with suffix appenders overlap with original passed column headers (which may result in an unintentional column overwrite in some cases)
- Results of validation returned in postprocess_dict['miscparameters_results']['columnoverlap_valresults']
- (False is good)
- Note that user is not expected to encounter this scenario in common practice, just trying to ensure robust to edge case
- Added some clarification to READ ME associated with passing Numpy Arrays - specifically that recomended practice is for any passed label column in train set be positioned as the final column of the set to ensure consistent column headers for test set (since test set may be passed without label column)
- Added option to automunge(.) parameter labels_column to pass as True in order to indicate that final column of the set is intended as labels_column (previously was only available for single column sets)

- Expanded range of parameter validations with new validation functions check_am_miscparameters and check_pm_miscparameters
- Now passed parameters with fixed set or fixed range of values are all validated to ensure legal entries
- Note that parameter validation results for automunge(.) are returned as postprocess_dict['miscparameters_results'] and the results for postmunge(.) are returned as postreports_dict['pm_miscparameters_results']
- A few clarifications added to parameter descriptions in READ ME

- Was working on some new passed parameter validations and realized had gotten mixed up in a prior validation between transformdict and processdict. No biggie, found and fixed.
- to be clear, transformdict contains the family tree entries for each root category, processdict contains the corresonding transformation functions and column properties associated with each category used as an entry in a family tree
- added a few images to the READ ME

- found and fixed small code typo in nmr7 and nmc7 transforms
- removed a redundant shuffle operation from postmunge feature importance derivation

- new datetime binned aggregations wkds, wkdo, mnts, mnto
- wkds and mnts are one-hot encoded days of week and months of year
- wkdo and mnto are comparable ordinal encoded sets
- new start/end parameters accepted for bshr (boolean marker for whether a time falls within business hours, which default to 9-5)
- new cap or floor parameters accepted for mnmx (min-max scaling)
- where default of False means no cap/floor, True means cap/floor set based on training data max/min, otherwise passed values put limit to the scaling range
- noting that if cap < max then max reset to cap, and if floor > min then min reset to floor
- please note that the wkds transform was inspired by a comment in a prerelease chapter for the "Deep Learning for Coders with fastai and PyTorch" book being written by Jeremy Howard and Sylvian Gugger
- removed amateurish relic from an early draft associated with global fillna() application