[BUG] ENN, RENN and AlKNN train KNN with different K from init param

[solegalli]

Describe the bug

ENN and RENN accept integers in the init parameter n_neighbors, to indicate the number of neighbours for the KNN algorithm
The default is 3. However, a 4 KNN is trained instead.

Steps/Code to Reproduce

import pandas as pd
from sklearn.datasets import make_classification
from imblearn.under_sampling import EditedNearestNeighbours

def make_data(sep):
    
    # returns arrays
    X, y = make_classification(n_samples=1000,
                           n_features=2,
                           n_redundant=0,
                           n_clusters_per_class=1,
                           weights=[0.99],
                           class_sep=sep,# how separate the classes are
                           random_state=1)
    
    # trasform arrays into pandas df and series
    X = pd.DataFrame(X, columns =['varA', 'varB'])
    y = pd.Series(y)
    
    return X, y

X, y = make_data(sep=2)

enn = EditedNearestNeighbours(
    sampling_strategy='auto',  
    n_neighbors=3, **# I expect a 3 KNN**
    kind_sel='all', 
    n_jobs=4,
)  

X_resampled, y_resampled = enn.fit_resample(X, y)

enn.nn_
>>> NearestNeighbors(n_jobs=4, n_neighbors=4)

Expected Results

enn.nn_

NearestNeighbors(n_jobs=4, n_neighbors=3)

Actual Results

NearestNeighbors(n_jobs=4, n_neighbors=4)

[glemaitre]

This is not a bug here. The thing is that a data sample is nearest neighbours of itself. Thus to respect the algorithm you need to add +1.

@chkoar is probably when explaining that we should not expose the internal in this case.

[solegalli]

Yes, I realised that after some digging of the source code :_(.

I am updating the docstrings in #850. Should I try to clarify somehow?