-
Notifications
You must be signed in to change notification settings - Fork 0
/
s-knn.py
226 lines (182 loc) · 7.44 KB
/
s-knn.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
# -*- coding: utf-8 -*-
#
# s-knn.py
#
# Created on 3/27/2020.
# Authors: Nikil Roashan Selvam, Varun Sivashankar.
import matplotlib as mpl
import matplotlib.pyplot as plt
import timeit
from sklearn.neighbors import KNeighborsClassifier
from sklearn import metrics
from sklearn.model_selection import KFold
import pandas as pd
import numpy as np
import sklearn
#Importing the data
def load_data(name):
if name=='iris':
iris = 'http://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data'
df = pd.read_csv(iris, sep=',')
attributes = ["sepal_length", "sepal_width", "petal_length", "petal_width", "class"]
df.columns = attributes
read_in_array = df.to_numpy()
for i in range(len(read_in_array)):
if read_in_array[i][4] == 'Iris-setosa':
read_in_array[i][4] = np.int64(0)
elif read_in_array[i][4] == 'Iris-versicolor':
read_in_array[i][4] = np.int64(1)
elif read_in_array[i][4] == 'Iris-virginica':
read_in_array[i][4] = np.int64(2)
else:
print("error")
return read_in_array
elif name=='balance-scale':
df = pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/balance-scale/balance-scale.data')
read_in_array = df.to_numpy()
one_hot = np.zeros((len(read_in_array),3))
for i in range(len(read_in_array)):
if read_in_array[i][0] == 'L':
one_hot[i][0] = 1
elif read_in_array[i][0] == 'B':
one_hot[i][1] = 1
elif read_in_array[i][0] == 'R':
one_hot[i][2] = 1
read_in_array = np.append(one_hot,read_in_array[:,1:],axis=1)
return read_in_array
elif name=='yeast':
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/yeast/yeast.data', delim_whitespace=True)
read_in_array = df.to_numpy()
read_in_array=read_in_array[:,1:]
yeast_labels={}
for idx,yeast_class in enumerate(read_in_array[:,-1]):
if yeast_class not in yeast_labels:
yeast_labels[yeast_class]=np.int64(len(yeast_labels))
read_in_array[idx][-1]=yeast_labels[yeast_class]
return read_in_array
elif name=='blood':
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/blood-transfusion/transfusion.data')
read_in_array = df.to_numpy()
return read_in_array
elif name=='haberman':
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/haberman/haberman.data')
read_in_array = df.to_numpy()
return read_in_array
elif name=='ion':
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/ionosphere/ionosphere.data')
read_in_array = df.to_numpy()
for i in range(len(read_in_array)):
if read_in_array[i][-1] == 'b':
read_in_array[i][-1] = 0
elif read_in_array[i][-1] == 'g':
read_in_array[i][-1] = 1
return read_in_array
elif name=='red-wine':
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv', sep=';')
read_in_array = df.to_numpy()
return read_in_array
else:
raise ValueError("Invalid dataset name: ", name)
def scaledKNN(data,num_epochs=1, alpha_0=1e-6, start_from_ones=False):
data_copy = np.copy(data)
n = len(data_copy[0])-1
#create dictionary of class against indices
class_to_inds = {}
for i in range(len(data_copy)):
c = data_copy[i][-1]
if c in class_to_inds:
class_to_inds[c].append(i)
else:
class_to_inds[c] = [i]
class_to_dists = {}
dists = np.zeros((1,n))
for c in class_to_inds:
temp = np.zeros((1,n))
inds = class_to_inds[c]
for i in range(len(inds)):
point1 = data_copy[i][:-1]
for j in range(i+1,len(inds)):
point2 = data_copy[j][:-1]
temp = temp + (point1-point2)**2
class_to_dists[c] = temp
dists = dists + temp
dists = dists.T
#initialize weights from uniform distribution between 0 and 1
if start_from_ones:
weights = np.ones((n,1))
else:
weights = np.random.rand(n,1)
cost = 0
old_cost = None
grad = np.zeros((n,1))
beta = 0.9 # momentum parameter
alpha = alpha_0
num_epoch = 0
while num_epoch < num_epochs:
cost = np.sum(np.multiply(weights**2,dists)) + np.sum(1/weights)
new_grad = 2 * np.multiply(weights,dists) - 1/weights**2
grad=(1-beta)*grad + beta*new_grad
weights = weights - alpha * grad
for i in range(len(weights)):
if weights[i] < 0: weights[i] = -weights[i]
weights = n * weights/np.sum(weights)
if old_cost != None and old_cost < cost:
alpha = alpha_0/(1+20*num_epoch)
old_cost = cost
num_epoch += 1
return weights, cost
# MAIN SCRIPT
datasets = ["iris", "balance-scale", "yeast", "blood", "haberman","ion","red-wine"]
seeds = [1,12,123,1234,12345,123456,1234567,12345678,123456789]
alphas = {'iris': 1e-08, 'balance-scale': 1e-05, 'yeast': 1e-07, 'blood': 1e-08, 'haberman': 0.001, 'ion': 1e-08, 'red-wine': 1e-06}
num_epochs = 6000
for dataset in datasets:
print("\t\t--Dataset: ", dataset)
data = load_data(dataset)
data = np.array(data, dtype=np.float64)
for seed in seeds:
print("\t\t--Seed: ", seed)
#create train-test split
np.random.seed(seed)
np.random.shuffle(data)
split = int(0.8*len(data))
train, test = data[:split,:], data[split:,:]
# normalize data
stds = np.std(train, axis = 0)
means = np.mean(train, axis = 0)
n = len(data[0]) - 1
epsilon = 10**(-8)
for i in range(n):
stds[i] = (stds[i]**2 + epsilon)**0.5
for i in range(n):
train[:,i] = (train[:,i] - means[i]) / stds[i]
test[:,i] = (test[:,i] - means[i]) / stds[i]
alpha = alphas[dataset]
weights,_ = scaledKNN(train,num_epochs=num_epochs,alpha_0=alpha,start_from_ones=True)
weights = np.array(weights).T * np.identity(len(weights))
#Get Results
train_X = train[:,0:-1]
train_y = list(train[:,-1])
test_X = test[:,0:-1]
test_y = list(test[:,-1])
cur_results = []
#Vanilla kNN
clf = KNeighborsClassifier(n_neighbors=5)
clf.fit(train_X,train_y)
train_ypred = clf.predict(train_X)
acc = metrics.accuracy_score(train_y, train_ypred, normalize=True)
print('Vanilla kNN:\t-- train acc %.3f' % acc)
test_ypred = clf.predict(test_X)
acc = metrics.accuracy_score(test_y, test_ypred, normalize=True)
print('Vanilla kNN:\t-- test acc %.3f' % acc)
#Scaled kNN
clf = KNeighborsClassifier(n_neighbors=5)
train_X = np.dot(train_X, weights)
test_X = np.dot(test_X, weights)
clf.fit(train_X,train_y)
train_ypred = clf.predict(train_X)
acc = metrics.accuracy_score(train_y, train_ypred, normalize=True)
print('Scaled kNN:\t-- train acc %.3f' % acc)
test_ypred = clf.predict(test_X)
acc = metrics.accuracy_score(test_y, test_ypred, normalize=True)
print('Scaled kNN:\t-- test acc %.3f' % acc)