SMOTETomek with SMOTE variants

[pckroon]

Description

Hi! First off, I know very little about machine learning in general, and imbalanced machine learning in particular, so I don't know if this will make much sense.
The problem I encountered is that I can not use combine.SMOTETomek with e.g. SVMSMOTE.

Steps/Code to Reproduce

import numpy as np
from imblearn.combine import SMOTETomek
from imblearn.over_sampling import SVMSMOTE

sampler = SMOTETomek(smote=SVMSMOTE())
sampler.fit_resample(np.arange(10).reshape(5, -1), np.arange(5))

Expected Results

A SMOTETomek sampler that uses SVMSMOTE for oversampling.

Actual Results

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/.../.virtualenvs/cartographer/lib/python3.6/site-packages/imblearn/base.py", line 84, in fit_resample
    output = self._fit_resample(X, y)
  File "/home/.../.virtualenvs/cartographer/lib/python3.6/site-packages/imblearn/combine/_smote_tomek.py", line 139, in _fit_resample
    self._validate_estimator()
  File "/home/.../.virtualenvs/cartographer/lib/python3.6/site-packages/imblearn/combine/_smote_tomek.py", line 117, in _validate_estimator
    'Got {} instead.'.format(type(self.smote)))
ValueError: smote needs to be a SMOTE object.Got <class 'imblearn.over_sampling._smote.SVMSMOTE'> instead.

Versions

>>> import platform; print(platform.platform())
Linux-4.15.0-54-generic-x86_64-with-Ubuntu-18.04-bionic
>>> import sys; print("Python", sys.version)
Python 3.6.7 (default, Oct 22 2018, 11:32:17) 
[GCC 8.2.0]
>>> import numpy; print("NumPy", numpy.__version__)
NumPy 1.16.2
>>> import scipy; print("SciPy", scipy.__version__)
SciPy 1.2.1
>>> import sklearn; print("Scikit-Learn", sklearn.__version__)
Scikit-Learn 0.21.2
>>> import imblearn; print("Imbalanced-Learn", imblearn.__version__)
Imbalanced-Learn 0.5.0

[hayesall]

Thanks for the question @pckroon! Currently the smote= parameter is used for passing a SMOTE object with parameters that are different from the defaults.

And it looks like the error is raised here:

imbalanced-learn/imblearn/combine/_smote_tomek.py

Lines 112 to 117 in 12b2e0d

	if self.smote is not None:
	if isinstance(self.smote, SMOTE):
	self.smote_ = clone(self.smote)
	else:
	raise ValueError('smote needs to be a SMOTE object.'
	'Got {} instead.'.format(type(self.smote)))

This should probably be adapted to accept SVMSMOTE and the other SMOTE variants as well.

[pckroon]

Changing

imbalanced-learn/imblearn/combine/_smote_tomek.py

Line 113 in 12b2e0d

if isinstance(self.smote, SMOTE):

for if instance(self.smote, BaseSMOTE): should do it code-wise (along with changing the corresponding import statement), but I don't know if there's some deeper reason why this would or would not be a bad idea.

PS. Is there any particular reason why ADASYN wouldn't work in this context?

[chkoar]

@pckroon basically by using the Pipeline object you could chain whatever samplers you want.

[pckroon]

Ah ok :)
I thought there was more going on than just calling one after the other.
In that case I would suggest removing/deprecating the SMOTEENN and SMOTETomek objects altogether, and in the combine docs write a little bit about using a pipeline to chain them. Currently it looks as if the combinations SMOTE+ENN and SMOTE+TomekLinks are special.

[glemaitre]

It is a bit easier to discover these samplers if you come from the literature.
You don't need to know about the internal and that it corresponds to make a pipeline.
If you read the paper then it is true that it is a bit overkill.