SMOTENC fails when all features are categorical

[linziheng]

I have a dataset with 4 categorical features and no continuous feature. When I run it on smote-nc, it throws an error saying

ValueError: Found array with 0 feature(s) (shape=(114965, 0)) while a minimum of 1 is required

Is there a way for smote-nc to work on dataset like this where features are all categorical?

Thanks!

[glemaitre]

Can you provide a stand-alone script to reproduce the problem ? Please read https://stackoverflow.com/help/mcve.
Also for future reference, can you paste the output of:

import platform; print(platform.platform())
import sys; print("Python", sys.version)
import numpy; print("NumPy", numpy.__version__)
import scipy; print("SciPy", scipy.__version__)
import sklearn; print("Scikit-Learn", sklearn.__version__)

[miguelcorrea]

I've run into the same issue. However, on closer inspection this seems to be by design (see previous issue and original paper). I'd suggest this should be made clearer in the documentation to prevent confusion.

[YunqiuXu]

Same issue, it seems that if I have 100 features, this extimator accepts 99 features at most, regardless of the indices

[glemaitre]

Please reopen an issue with a minimal reproducible example. We cannot fix the issue without it.

[mariolovric]

CHANGE LINE 17 TO MAKE ALL FEATURES CATEGORICAL

#create classification set with mixed data
from sklearn.datasets import make_classification
from imblearn.over_sampling import SMOTENC
import pandas as pd
def make_mix_class(n_classes=2, n_features=20,n_samples=300):
    X, y = make_classification(
        n_classes=n_classes, class_sep=1, weights=[.9,.1],
        n_informative=10, n_redundant=1, flip_y=0,
        n_features=n_features, n_clusters_per_class=1,
        n_samples=n_samples, random_state=10
    )

    X=pd.DataFrame(X)
    y=pd.Series(y)
    col_list = ['s' + str(x) for x in range(0,n_features)]
    X.columns=col_list
    for i,col in enumerate(col_list):
        if i >= 1: #change here to 0 for all categorical
          
            X[col]=pd.cut(X[col], bins=3, labels=[1,2,3])


    return X,y

X,y = make_mix_class(n_classes=2, n_features=20,n_samples=100)
X.apply(lambda x: x.value_counts() )




indices = [i for i in range(len(X.columns)) if X.iloc[:,i].isin([1,2,3]).all()]
X_resampled, y_resampled = SMOTENC(categorical_features=indices, sampling_strategy='not majority', n_jobs=6, random_state=42).fit_resample(X, y)
print('y: \n',pd.Series(y).value_counts(),'\n\n','y resampled: \n',pd.Series(y_resampled).value_counts())

[mariolovric]

Please reopen an issue with a minimal reproducible example. We cannot fix the issue without it.

please re-open, I have added code for testing

[Yevgnen]

Same issue here.

[popkir]

same here

[yenerhelvacioglu]

I am having the same issue

[ShaneTian]

Same issue here @glemaitre

[congbo]

Same issue here @glemaitre

[chkoar]

This is obviously a bug. If all features are categorical the sampler fails. Although, there is an on going PR that supports purely categorical data.

Using the code provided by @mariolovric a pure categorical dataset can be created

import pandas as pd
from sklearn.datasets import make_classification


n_features = 20

X, y = make_classification(
    n_classes=2,
    class_sep=1,
    weights=[0.9, 0.1],
    n_informative=10,
    n_redundant=1,
    flip_y=0,
    n_features=n_features,
    n_clusters_per_class=1,
    n_samples=300,
    random_state=10,
)


X = pd.DataFrame(X)
y = pd.Series(y)
col_list = ["s" + str(x) for x in range(0, n_features)]
X.columns = col_list
for i, col in enumerate(col_list):
    X[col] = pd.cut(X[col], bins=3, labels=[1, 2, 3])

As a workaround, the following helper function should do the trick

from imblearn.over_sampling import SMOTENC

def sample(X, y, sampler=None):
    if sampler is None:
        sampler = SMOTENC(categorical_features=[], random_state=42)
    X["temp"] = 0
    n_features = X.shape[1] - 1
    indices = range(n_features)
    sampler.set_params(categorical_features=indices)
    X_resampled, y_resampled = sampler.fit_resample(X, y)
    del X_resampled["temp"]
    return X_resampled, y_resampled

X_resampled, y_resampled = sample(X, y)