Automunge

- new automunge and postmunge parameter LSfit
- LSfit removes assumption of equal distribution of labels for smoothing parameter K in label smoothing to a fitted K tailored to activation ratios associated with each label category
- Thus LSfit introduces a little more intelligence into the Label Smoothing equation by way of creating a parameterized smoothing factor K as a function of the activation column and the target column associated with each cell
- LSfit defaults to False for prior assumption of even distribution of label classes, conducts fitting operation when passed as True
- I'll have to put some thought into it but am currently undeciuded if LSfit has benefit in cases when conducting oversampling of training data for class imbalance in labels via the TrainLabelFreqLevel option, it might still.

- a user can now pass True to automunge(.) parameters LabelSmoothing_test and LabelSmoothing_val to consistently encode as parameter LabelSmoothing_train
- a user can now pass True to postmunge(.) parameter LabelSmoothing to consistently encode as automunge(.) parameter LabelSmoothing_train accessed from the postprocess_dict

- new edge case for label smoothing
- if labels category evaluated as 'bnry' and label smoothing desired
- reset labels category to 'text' (one hot encoding)
- a reminder 'bnry' is the single column encoding of binary variables

- new automunge parameters LabelSmoothing_train / LabelSmoothing_test / LabelSmoothing_val
- new postmunge parameter LabelSmoothing
- note that Label Smoothing as implemented still supports oversampling preparation via TrainLabelFreqLevelizer
- Label Smoothing refers to the regularization tactic of transforming boolean encoded labels from 1/0 designations to some mix of reduced/increased threshold - for example passing the float 0.9 would result in the conversion from the set 1/0 to 0.9/, where is a function of the number of cateogries in the label set - for example for a boolean label it would convert 1/0 to 0.9/0.1, or for the one-hot encoding of a three label set it would be convert 1/0 to 0.9/0.05. Hat tip for the concept to "Rethinking the Inception Architecture for Computer Vision" by Szegedy et al.

- fixed a small bug with functionality rolled out in 2.95 (found an edge case)
- fixed a small bug for assignparam associated with feature importance evaluation

- new option for user to overwrite transformation function parameters on all columns
- without need to assign them individually
- by passing entries in assignparam under 'default_assignparam'
- further details of logistics provided in READ ME
- note that thisz method only overwrites the default parameters for those not otherwise specified