实现PLA和Pocket算法
import sys
import numpy as np
import random as rd
def loadfile(file_path):
fil = open(file_path)
lines = fil.readlines()
num = len(lines)
X = np.zeros((num,5))
Y = np.zeros((num,1))
index = 0
for line in lines:
items = line.strip().split('\t')
X[index][1:5] = np.array([float(i) for i in items[0].strip().split()])[:]
X[index][0] = 1
Y[index][0] = float(items[1])
index += 1
return X,Y
def pla_error_rate(features, labels, w):
length = len(features)
wrong = 0
for i in range(length):
if labels[i][0] * (np.dot(features[i], w))[0] <= 0:
wrong += 1
return float(wrong)/float(length)
def pla_pocket(features, labels, index_array, max_times, rate = 1):
w = np.zeros((5,1))
w_pocket = np.zeros((5,1))
num = len(features)
flag = 1
index = 0
count = 0
while(flag):
features_index = index_array[index]
if labels[features_index][0] * np.dot(features[features_index], w)[0] <= 0:
w = w + labels[features_index][0] * rate * np.mat(features[features_index]).T
count += 1
if pla_error_rate(features, labels, w) < pla_error_rate(features, labels, w_pocket):
w_pocket = w
if count >= max_times:
flag = 0
elif index >= num - 1:
index = 0
else:
index += 1
return w, w_pocket
def pla(features, labels, rate = 1):
w = np.zeros((5,1))
num = len(features)
flag = 1
index = 0
good_items = 0
count = 0
while(flag):
if labels[index][0] * np.dot(features[index], w)[0] <= 0:
w = w + labels[index][0] * rate * np.mat(features[index]).T
good_items = 0
count += 1
else:
good_items += 1
if good_items >= num:
flag = 0
elif index >= num - 1:
index = 0
else:
index += 1
return count
def pla_fix(features, labels, index_array, rate = 1):
w = np.zeros((5,1))
num = len(features)
flag = 1
index = 0
good_items = 0
count = 0
while(flag):
features_index = index_array[index]
if labels[features_index][0] * np.dot(features[features_index], w)[0] <= 0:
w = w + labels[features_index][0] * rate * np.mat(features[features_index]).T
good_items = 0
count += 1
else:
good_items += 1
if good_items >= num:
flag = 0
elif index >= num - 1:
index = 0
else:
index += 1
return count
if __name__ == '__main__':
### homework0 15
"""
(X,Y) = loadfile('data.txt')
print(pla(X, Y))
"""
### homework0 16
"""
(X,Y) = loadfile('data.txt')
update_array = []
for i in range(2000):
index_array = [j for j in range(400)]
rd.shuffle(index_array)
tmp = pla_fix(X, Y, index_array)
update_array.append(tmp)
print(np.mean(update_array))
"""
### homework0 17
"""
(X,Y) = loadfile('data.txt')
update_array = []
for i in range(200):
index_array = [j for j in range(400)]
rd.shuffle(index_array)
tmp = pla_fix(X, Y, index_array, rate = 0.5)
update_array.append(tmp)
print(update_array)
"""
### homework0 18
(X,Y) = loadfile('pocket_data.txt')
(X_test,Y_test) = loadfile('pocket_test.txt')
error_rate_array = []
for i in range(200):
index_array = [j for j in range(len(X))]
rd.shuffle(index_array)
(w, w_100) = pla_pocket(X, Y, index_array, 100)
error_rate_array.append(pla_error_rate(X_test, Y_test, w_100))
print(np.mean(error_rate_array))
references:
https://blog.csdn.net/rikichou/article/details/78226036
https://blog.csdn.net/hulingyu1106/article/details/51212632
网友评论