pr_exp1/spect/pla_np.py

#!/usr/bin/env python3
import numpy as np

c = 0.1
LIMIT = 100
print_epoches = [5, 10, 100, 1000]
DATA_SLICE = 5
NUM_CLASS = 1
INPUT_FEATURE = 22

def get_one_hot(targets, nb_classes):
    res = np.eye(nb_classes)[np.array(targets).reshape(-1)]
    return res.reshape(list(targets.shape)+[nb_classes])

def load(file):
    raw_data = np.loadtxt(file, delimiter=',')
    data_size = len(raw_data)
    np.random.shuffle(raw_data)
    ret = np.split(raw_data, [1,], axis=1)
    return ret

train_pieces = load('SPECT.train')
train_label = train_pieces[0]
train_data = train_pieces[1]
train_label = train_label.flatten().astype(int)


val_pieces = load('SPECT.test')

val_label = val_pieces[0]
val_data = val_pieces[1]
val_label = val_label.flatten().astype(int)

v = np.array(val_data, dtype='float32')
v = np.concatenate((v, np.ones((len(v), 1))), axis=1)

def pla(train_data, train_label, val_data, val_label, picked_class):
    x = np.array(train_data, dtype='float32')
    x = np.concatenate((x, np.ones((len(x), 1))), axis=1)

    for i in range(len(x)):
        if train_label[i] != picked_class:
            x[i] *= -1
    w = np.random.rand(INPUT_FEATURE + 1)
    for j in range(LIMIT):
        flag = False
        for i in x:
            z = sum(i * w)
            if z <= 0:
                w = w + c * i
                flag = True
        if not flag:
            print('')
            break
        #if j in print_epoches:
        print(j)
        print('train acc:', np.sum(np.sum(x * w, axis=1) > 0) / len(x))
        print('val acc:', np.sum((np.sum(v * w, axis=1) > 0) == (val_label == picked_class)) / len(v))
    return w


w = [pla(train_data, train_label, val_data, val_label, i) for i in range(NUM_CLASS)]
pred = [np.sum(v * w[i], axis=1) > 0 for i in range(NUM_CLASS)]
TP = sum(pred[0][i] == 1 and val_label[i] == 1 for i in range(len(val_label)))
FP = sum(pred[0][i] == 1 and val_label[i] == 0 for i in range(len(val_label)))
FN = sum(pred[0][i] == 0 and val_label[i] == 1 for i in range(len(val_label)))
P = TP / (TP + FP)
R = TP / (TP + FN)
print('val acc:', sum([all([pred[j][i] == (j == val_label[i]) for j in range(NUM_CLASS)]) for i in range(len(val_label))]) / len(val_label))
print('F1', 2 * P * R / (P + R))