[Feat]: Log1p transformer

[TremaMiguel]

Is your feature request related to a problem? Please describe.

When you want to apply a logarithmic transformation and data contains a zero value, calculating log(1+x) is more convenient.

Describe the solution you'd like

This could be another argument in LogTransformer that checks for zeros and apply log(1 + x)

from feature_engine import transformation as vt

tf = vt.LogTransformer(variables = ['LotArea', 'GrLivArea'], check_zeros=True)

under the hood basically applies

if check_zeros:
    log_1p = True if (X[self.variables_] == 0).any().any() else False

or it can be a separate transformer

from feature_engine import transformation as vt

tf = vt.Log1PTransformer(variables = ['LotArea', 'GrLivArea'], check_zeros=True)

[solegalli]

What would this transformer do if check_zeroes is True? add 1 to the variable and then apply the log?

I am aware that this is done quite frequently. I just think it is dangerous to add 1 to a variable without really inspecting the variable nature that I am kind of reluctant to encourage its use by implementing it in a transformer.

Also why +1 and not +10?

[TremaMiguel]

I think it would be better for the user to decide which constant C to add and handle it as a separate transformer. For example,

class LogCpTransformer(BaseNumericalTransformer):
    """
    The LogCpTransformer() applies the transformation log( x + C) where C is a positive constant to 
    the input variable. Adding a C constant helps apply log transformation to a variable having zero
    or negative values. 
    
    The LogTransformer() only works with numerical non-negative values. If the variable
    contains a zero or a negative value after adding a constant C, the transformer will return an error.

[glevv]

In LogCpTransformer we can just set C=0 by default and also add option 'auto' which will do something like that C = abs(min(x)) + 1. This will allow us to use log transformation on negative values.
Also this one is bad

If the variable contains a zero or a negative value after adding a constant C, the transformer will return an error.

because log(0+1) = 0, so you will raise error in valid case. The reason behind log1p is to avoid infs and nans that will arise in log(0) case, not in avoiding zeros.

[solegalli]

If the variable contains a zero or a negative value after adding a constant C, the transformer will return an error.

that line checks the variable before the transformation.

[TremaMiguel]

@solegalli agree on developing this feature to open a PR ?

[solegalli]

go for it :)

[solegalli]

closed in #281