LogEndogTransformer doesn't work in pipelines

[jseabold]

Describe the bug

When cloning a transformer, it gets the parameter names from the __init__. When using LogEndogTransformer, you never set lmbda, you set lmbda2 for the Box-Cox transformer, so when you clone the transformer, it thinks the values of lmbda is 0 even though you may have set it to something else.

To Reproduce
Steps to reproduce the behavior:

In [3]: from pmdarima.preprocessing.endog import LogEndogTransformer

In [4]: from sklearn.base import clone

In [5]: LogEndogTransformer(lmbda=10).fit_transform([0, 10])
Out[5]: (array([2.3026, 2.9957]), None)

In [6]: clone(LogEndogTransformer(lmbda=10)).fit_transform([0, 10])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-6-360cffae9200> in <module>
----> 1 clone(LogEndogTransformer(lmbda=10)).fit_transform([0, 10])

~/src/pmdarima/pmdarima/preprocessing/base.py in fit_transform(self, y, X, **kwargs)
     55         """
     56         self.fit(y, X, **kwargs)  # TODO: eventually do not pass kwargs to fit
---> 57         return self.transform(y, X, **kwargs)
     58
     59     @abc.abstractmethod

~/src/pmdarima/pmdarima/preprocessing/endog/log.py in transform(self, y, X, **transform_kwargs)
     73             The exog array
     74         """
---> 75         return super().transform(y, X, **transform_kwargs)
     76
     77     def inverse_transform(self, y, X=None, **kwargs):  # TODO: kwargs go away

~/src/pmdarima/pmdarima/preprocessing/endog/boxcox.py in transform(self, y, X, **kwargs)
    123             msg = "Negative or zero values present in y"
    124             if action == "raise":
--> 125                 raise ValueError(msg)
    126             elif action == "warn":
    127                 warnings.warn(msg, UserWarning)

ValueError: Negative or zero values present in y

In [7]: clone(LogEndogTransformer(lmbda=10))
Out[7]: LogEndogTransformer(floor=1e-16, lmbda=0, neg_action='raise')

Versions

I'm running a development version to work around a few issues such as this one. Tag is 23d4d0d.

System:
    python: 3.7.6 | packaged by conda-forge | (default, Mar  5 2020, 15:03:29)  [Clang 9.0.1 ]
executable: /Users/skipper/.miniconda3/bin/python
   machine: Darwin-19.6.0-x86_64-i386-64bit

Python dependencies:
        pip: 20.3.3
 setuptools: 49.6.0.post20201009
    sklearn: 0.22
statsmodels: 0.12.1
      numpy: 1.19.4
      scipy: 1.4.1
     Cython: 0.29.17
     pandas: 1.1.3
     joblib: 1.0.0
   pmdarima: 0.0.0

If I can work out a patch quickly, I'll make a PR, but not sure how to work around this off the top of my head.

[jseabold]

Well, this is the easiest workaround. Not thoroughly tested.

diff --git a/pmdarima/preprocessing/endog/log.py b/pmdarima/preprocessing/endog/log.py
index 5ad8e8ed..495ed5cb 100644
--- a/pmdarima/preprocessing/endog/log.py
+++ b/pmdarima/preprocessing/endog/log.py
@@ -101,3 +101,25 @@ class LogEndogTransformer(BoxCoxEndogTransformer):
             The inverse-transformed exogenous array
         """
         return super().inverse_transform(y, X, **kwargs)
+
+    def get_params(self, deep=True):
+        """
+        Get parameters for this estimator.
+
+        Parameters
+        ----------
+        deep : bool, default=True
+            If True, will return the parameters for this estimator and
+            contained subobjects that are estimators.
+
+        Returns
+        -------
+        params : mapping of string to any
+            Parameter names mapped to their values.
+        """
+        # this method is shadowed because we shadow the the lmbda parameter
+        # name. Instead of setting lmbda, this transformer stores the parameter
+        # as lmbda2
+        params = super().get_params(deep=deep)
+        params['lmbda'] = self.lmbda2
+        return params

[tgsmith61591]

Thanks @jseabold great report. We will get this fixed up