DatasetMembershipTransformer.py

Go to the documentation of this file.

from ._CustomTransformer import _CustomTransformer
 
from sklearn import preprocessing
import numpy as np 
 
 
class DatasetMembershipTransformer(_CustomTransformer):
    ''' Appends a one-hot vector of features to each sample, indicating dataset membership  '''
 
    def __init__(self, datasets, *args, **kwargs):
        self.locs   = datasets.flatten()[:, None]
        self.ohe    = preprocessing.OneHotEncoder(sparse=False).fit(self.locs)
        self.n_sets = len(np.unique(self.locs))
 
    def _transform(self, X, *args, idx=None, zeros=False, **kwargs):
        # if X is not the same shape as locs, and idx=None, we just append zeros
        # otherwise, we append the appropriate one-hot vector corresponding to a 
        # sample's dataset membership
        if ((X.shape[0] == len(self.locs)) or (idx is not None)) and not zeros:
            idx = idx or slice(None)
            loc = self.locs[idx]
            return np.append(X, self.ohe.transform(loc), 1)
        return np.append(X, np.zeros((len(X), self.n_sets)), 1)
 
    def _inverse_transform(self, X, *args, **kwargs):
        return X[:, :-self.n_sets]