Error with under_sampling.CondensedNearestNeighbour

[WafaMribah]

I am trying to balance my data by reducing the number of samples of the class 0 but every time I specify the ratio ( via the sampling_strategy parameter, I get the following error :

Code

import matplotlib.pyplot as plt
import numpy as np

from sklearn.decomposition import PCA
from imblearn.under_sampling import (CondensedNearestNeighbour,
                                     EditedNearestNeighbours, AllKNN)

print(__doc__)

pca = PCA(n_components=2)

sampling_strategy = {0: 3500, 1:171}
cnn = CondensedNearestNeighbour(sampling_strategy)
X_resampled, y_resampled = cnn.fit_sample(X_train, Y_train)

ValueError: 'sampling_strategy' as a dict for cleaning methods is not supported. Please give a list of the classes to be targeted by the sampling. 

Remark

If the sampling strategy parameter is indeed removed as mentioned in the documentation then I would like to know if there is another way to specify the ratio.

Versions

Linux-4.9.119-44.140.amzn1.x86_64-x86_64-with-glibc2.9
Python 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 18:10:19)
[GCC 7.2.0]
NumPy 1.17.0
SciPy 1.3.1
Scikit-Learn 0.22.1
Imbalanced-Learn 0.6.1

[glemaitre]

You cannot specify the ratio with methods which are cleaning methods as specified in the user guide, in the API documentation, and by the error. Cleaning methods are method which will automatically clean the feature space and does not allow any user interaction.

cf. https://imbalanced-learn.org/stable/under_sampling.html#cleaning-under-sampling-techniques

[esaracin]

There appears to still be an issue with the sampling_strategy argument in CondensedNearestNeighbour and other Cleaning methods like TomekLinks.

According to the API, "when [sampling_strategy is a] callable, function taking y and returns a dict. The keys correspond to the targeted classes. The values correspond to the desired number of samples for each class."

However, doesn't seem to work, as

...
  def produce_sampling_strategy(target):
      num_minority_samples: int = sum(target)
      num_majority_samples_needed_overall: int = ceil(num_minority_samples / sampling_ratio)
      return {
          0: num_majority_samples_needed_overall,
          1: num_minority_samples
      }

  print(
      f"Before down-sampling, there are {len(target)} rows, "
      f"and the target breakdown is: {Counter(target)}."
  )
  tl = TomekLinks(sampling_strategy=produce_sampling_strategy)

  # Combine struct and text features. Before doing so, grab the number of structured features, so we can
  # split them after resampling is complete.
  training_features_to_downsample = torch.cat((struct, text), 1)
  X_res, y_res = tl.fit_resample(training_features_to_downsample, target)
...

fails with the following error:
ValueError: 'sampling_strategy' as a dict for cleaning methods is not supported. Please give a list of the classes to be targeted by the sampling.

[ramdhan1989]

is this solved? I have same problem