data_interpolation.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@author: Guansong Pang

Source code for the PReNet algorithm in KDD'23.
"""

import numpy as np
from scipy.sparse import vstack, csc_matrix
from sklearn.neighbors import KDTree

def inject_noise_sparse(seed, n_out):     
    rng = np.random.RandomState(42) 
    n_sample, dim = seed.shape
    swap_ratio = 0.05
    n_swap_feat = int(swap_ratio * dim)
    seed = seed.tocsc()
    noise = csc_matrix((n_out, dim))
    print(noise.shape)
    for i in np.arange(n_out):
        outlier_idx = rng.choice(n_sample, 2, replace = False)
        o1 = seed[outlier_idx[0]]
        o2 = seed[outlier_idx[1]]
        swap_feats = rng.choice(dim, n_swap_feat, replace = False)
        noise[i] = o1.copy()
        noise[i, swap_feats] = o2[0, swap_feats]
    return noise.tocsr()

def inject_noise(seed, n_out):     
    rng = np.random.RandomState(42) 
    n_sample, dim = seed.shape
    swap_ratio = 0.05
    n_swap_feat = int(swap_ratio * dim)
    noise = np.empty((n_out, dim))
    for i in np.arange(n_out):
        outlier_idx = rng.choice(n_sample, 2, replace = False)
        o1 = seed[outlier_idx[0]]
        o2 = seed[outlier_idx[1]]
        swap_feats = rng.choice(dim, n_swap_feat, replace = False)
        noise[i] = o1.copy()
        noise[i, swap_feats] = o2[swap_feats]
    return noise