smotenc_for_pandas.py

# https://imbalanced-learn.readthedocs.io/en/stable/generated/imblearn.over_sampling.SMOTENC.html

# SMOTENC (SMOTE) for Pandas DataFrame
# - this codes uses SMOTENC (imbalanced-learn library) for oversampling imbalanced data
# - it preserves dataframe object, columns names and dtypes

from imblearn.over_sampling import SMOTENC

def col_ins(ds, var):
  # column names to indices
  return [ds.columns.get_loc(col) for col in var]

def smotenc(X, y, cat_var_ins):
  sm = SMOTENC(random_state=42, categorical_features=cat_var_ins)
  return sm.fit_sample(X, y)

def df_smotenc(df, dep_var, cat_var):
  y = df[dep_var]
  X = df.drop(dep_var, axis=1)
  cat_var_ins = col_ins(X, cat_var)
  
  # smotenc
  X_res, y_res = smotenc(X, y, cat_var_ins)
  
  # back to DataFrame (SMOTENC uses numpy)
  X_res = pd.DataFrame(X_res, columns=X.columns)
  y_res = pd.DataFrame(y_res, columns=[dep_var])
  df_res = y_res.merge(X_res, left_index=True, right_index=True)

  # set dtypes (which are lost when SMOTENC uses numpy)
  df_res = df_res.astype((ds.dtypes))

  return df_res