Problem implementing a custom transformer for multilabel SMOTE

[romanwolf-git]

I am trying to implement a custom transformer for multilabel SMOTE in my pipeline. However, I constantly run into:
"All intermediate steps of the chain should be estimators that implement fit and transform or fit_resample (but not both) or be a string 'passthrough' 'MLSMOTE_resampler(n_samples=100)' (type <class 'upsampling_multilabel_data_with_mlsmote.MLSMOTE_resampler'>) doesn't). Here's my code for the class

import numpy as np
import pandas as pd
import random
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.datasets import make_classification
from sklearn.neighbors import NearestNeighbors


def get_tail_label(df: pd.DataFrame, ql=[0.05, 1.]) -> list:
    """
    Find the underrepresented targets.
    Underrepresented targets are those which are observed less than the median occurence.
    Targets beyond a quantile limit are filtered.
    """
    irlbl = df.sum(axis=0)
    irlbl = irlbl[(irlbl > irlbl.quantile(ql[0])) & ((irlbl < irlbl.quantile(ql[1])))]  # Filtering
    irlbl = irlbl.max() / irlbl
    threshold_irlbl = irlbl.median()
    tail_label = irlbl[irlbl > threshold_irlbl].index.tolist()
    return tail_label


def get_minority_samples(X: pd.DataFrame, y: pd.DataFrame, ql=[0.05, 1.]):
    """
    return
    X_sub: pandas.DataFrame, the feature vector minority dataframe
    y_sub: pandas.DataFrame, the target vector minority dataframe
    """
    tail_labels = get_tail_label(y, ql=ql)
    index = y[y[tail_labels].any(axis=1)].index.tolist()

    X_sub = X[X.index.isin(index)].reset_index(drop=True)
    y_sub = y[y.index.isin(index)].reset_index(drop=True)

    return X_sub, y_sub


def nearest_neighbour(X: pd.DataFrame, neigh) -> list:
    """
    Give index of 10 nearest neighbor of all the instance
    
    args
    X: np.array, array whose nearest neighbor has to find
    
    return
    indices: list of list, index of 5 NN of each element in X
    """
    nbs = NearestNeighbors(n_neighbors=neigh, metric='euclidean', algorithm='brute').fit(X)
    euclidean, indices = nbs.kneighbors(X)
    return indices


def MLSMOTE(X, y, n_sample, neigh=5):
    """
    Give the augmented data using MLSMOTE algorithm
    
    args
    X: pandas.DataFrame, input vector DataFrame
    y: pandas.DataFrame, feature vector dataframe
    n_sample: int, number of newly generated sample
    
    return
    new_X: pandas.DataFrame, augmented feature vector data
    target: pandas.DataFrame, augmented target vector data
    """
    indices2 = nearest_neighbour(X, neigh=5)
    n = len(indices2)
    new_X = np.zeros((n_sample, X.shape[1]))
    target = np.zeros((n_sample, y.shape[1]))
    for i in range(n_sample):
        reference = random.randint(0, n - 1)
        neighbor = random.choice(list(indices2[reference, 1:]))
        all_point = indices2[reference]
        nn_df = y[y.index.isin(all_point)]
        ser = nn_df.sum(axis=0, skipna=True)
        target[i] = np.array([1 if val > 0 else 0 for val in ser])
        ratio = random.random()
        gap = X.loc[reference, :] - X.loc[neighbor, :]
        new_X[i] = np.array(X.loc[reference, :] + ratio * gap)
    new_X = pd.DataFrame(new_X, columns=X.columns)
    target = pd.DataFrame(target, columns=y.columns)
    return new_X, target


class MLSMOTE_resampler(BaseEstimator, TransformerMixin):
    def __init__(self, n_samples=500, neigh=5):
        self.n_samples = n_samples
        self.neigh = neigh

    def _fit_resample(self, X, y):
        X = pd.DataFrame.sparse.from_spmatrix(X)
        X_sub, y_sub = get_minority_samples(X, y)
        X_res, y_res = MLSMOTE(X_sub, y_sub)
        X_con = pd.concat(X, X_res, ignore_index=True)
        y_con = pd.concat(y, y_res, ignore_index=True)
        return X_con, y_con

This is my pipeline:

from imblearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer
from sklearn.multioutput import MultiOutputClassifier

from ml_pipeline_preparation tokenize
from upsampling_multilabel_data_with_mlsmote import MLSMOTE

pipe = Pipeline([
    ('vect', CountVectorizer(tokenizer=tokenize)),
    ('tfidf', TfidfTransformer()),
    ('mlsmote', sampler()),
    ('clf', MultiOutputClassifier(RandomForestClassifier()))
])

[solegalli]

Could it be that the method _fit_resample() in MLSMOTE_resampler has a preceding underscore? Try removing the _.

[romanwolf-git]

@solegalli I tried it with and without still with the same result. The reason why I tried it with was just to follow the SMOTE class of imbalanced-learn.

[chkoar]

Not inheriting from the TransformerMixin and by implementing fit_resample, as @solegalli mentioned, will fix your problem.