iCROUST.py

# Creator: Atif Hassan
# email-id: atif.hassan@iitkgp.ac.in
# email-id: atif.hit.hassan@gmail.com

'''
Copyright [2019] [Atif Hassan]

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
'''

import numpy as np
import math
from sklearn.cluster import KMeans, MiniBatchKMeans
from sklearn.neighbors import NearestNeighbors

def icroust(X, Y, points_to_remove, n_neighbours, points_to_remove_at_a_time=4, num_clusters=10000, maj_class=0, min_class=1):
    X_minority = X[np.where(Y==min_class)[0]]
    X_majority = X[np.where(Y==maj_class)[0]]
    indices = list()
    if cluster_centers > len(X_minority):
        dists_cent = np.array([min([np.linalg.norm(i-j) for j in X_minority]) for i in X_majority])
    else:
        #Cluster minority samples
        kmeans = MiniBatchKMeans(n_clusters=num_clusters, max_iter=100, random_state=0).fit(X_minority)
        #1. Find centroids
        centroids = kmeans.cluster_centers_
        #2. Get distance of majority class samples from closest centroid
        dists_cent = np.array([min([np.linalg.norm(i-j) for j in centroids]) for i in X_majority])

    print("Points to remove = "+str(points_to_remove))
    #for k in range(0, points_to_remove, points_to_remove_at_a_time):
    for k in range(math.floor(points_to_remove/points_to_remove_at_a_time)):
        #3. Sort majority samples by above distance in ascending order (could be important samples. Sort in descending order ?)
        #4. Get average distance with k=3 nearest neighbour of majority samples
        neigh = NearestNeighbors(n_neighbors=n_neighbours+1, algorithm='kd_tree', n_jobs=-1).fit(X_majority)
        dist, inds = neigh.kneighbors(X_majority)
        dists_nn = np.array([sum(dist[i])/n_neighbours for i in range(len(X_majority))])
        samples = [[i, dists_cent[i]+dists_nn[i]] for i in range(len(X_majority))]
        samples.sort(key=lambda z: z[1])
        #6. Now remove top k majority samples
        indices+= [i[0] for i in samples[:points_to_remove_at_a_time]]
        X_majority = np.delete(X_majority, [i[0] for i in samples[:points_to_remove_at_a_time]], axis=0)
    X, Y = list(), list()
    for i in X_majority:
        X.append(i)
        Y.append(maj_class)
    for i in X_minority:
        X.append(i)
        Y.append(min_class)
    return np.array(X), np.array(Y)
    

    return X, Y