I want to create a function, that takes data instances, labels, and a target-proportion.
The function should determine the proportion of classes in the given dataset/labels, and resample the data into the given target proportion using either imblearn.over_sampling.SMOTE for classes that require over-sampling to reach its target-proportion, or the imblearn.under_sampling.RandomUnderSampler for classes that need be under-sampled to make it equal to target-proportion.

For example, given:

X, y = make_classification(n_samples=10000, n_features=10, n_classes=5,
                           n_informative=4, weights=[0.3,0.125,0.239,0.153,0.188])

X_train,X_test,y_train,y_test=train_test_split(X,y, random_state=42)

Initially, the proportion is:

class 0:  0.3, 
class 1: 0.125, 
class 2: 0.239, 
class 3: 0.153, 
class 4: 0.188

And we want to get the following target proportion:

class 0: 0.519 
class 1: 0.373  
class 2: 0.226
class 3: 0.053
class 4: 0.164

The function should determine when to use SMOTE or RandomUndersampler.

My code:

from imblearn.over_sampling import SMOTE
from imblearn.under_sampling import RandomUnderSampler
from collections import Counter

def resample_to_proportion(X, y, target_proportion):
    # Calculate the current proportion of each class
    class_counts = Counter(y)
    total_samples = len(y)
    current_proportion = {label: count / total_samples for label, count in class_counts.items()}

    # Initialize resampling strategies
    resampling_strategies = {}
    for label, target_prop in target_proportion.items():
        if target_prop > current_proportion[label]:
            resampling_strategies[label] = SMOTE(sampling_strategy=target_prop)
        elif target_prop < current_proportion[label]:
            resampling_strategies[label] = RandomUnderSampler(sampling_strategy=target_prop)

    # Resample each class based on the difference between current and target proportion
    X_resampled = []
    y_resampled = []
    for label, strategy in resampling_strategies.items():
        mask = y == label
        X_class = X[mask]
        y_class = y[mask]
        X_resampled_class, y_resampled_class = strategy.fit_resample(X_class, y_class)
        X_resampled.append(X_resampled_class)
        y_resampled.append(y_resampled_class)

    # Concatenate resampled data
    X_resampled = np.concatenate(X_resampled)
    y_resampled = np.concatenate(y_resampled)

    return X_resampled, y_resampled

But then:

target_proportion = {0: 0.519, 1: 0.373, 2: 0.226, 3: 0.053, 4: 0.164}

X_resampled, y_resampled = resample_to_proportion(X_train, y_train, target_proportion)

ValueError: The target 'y' needs to have more than 1 class. Got 1 class instead