# -*- coding: utf-8 -*-

"""

Created on Tue Jul 05 16:48:32 2016

@author: WuJianping

"""

import re

import matplotlib.pyplot as plt

def parser_log(logfile):

regex_iteration = re.compile('Iteration (\d+)')

regex_train_output_accuarcy = re.compile('Train net output #0: (\S+) = ([\.\deE+-]+)')

regex_train_output_loss = re.compile('Train net output #1: (\S+) = ([\.\deE+-]+)')

regex_test_output_accuarcy = re.compile('Test net output #0: (\S+) = ([\.\deE+-]+)')

regex_test_output_loss = re.compile('Test net output #1: (\S+) = ([\.\deE+-]+)')

regex_learning_rate = re.compile('lr = ([-+]?[0-9]*\.?[0-9]+([eE]?[-+]?[0-9]+)?)')

iteration = -1;

learning_rate = [[], []]

train_accuracy = [[], []]

train_loss = [[], []]

test_accuracy = [[], []]

test_loss = [[], []]

with open(logfile, "r") as fp:

for line in fp:

# print line

iteration_match = regex_iteration.search(line)

if iteration_match:

iteration = int(iteration_match.group(1))

if -1 == iteration:

continue;

# print iteration

learning_rate_match = regex_learning_rate.search(line)

if learning_rate_match:

lr = float(learning_rate_match.group(1))

# print lr

learning_rate[0].append(iteration)

learning_rate[1].append(lr)

train_accuracy_match = regex_train_output_accuarcy.search(line)

if train_accuracy_match:

trainacc = float(train_accuracy_match.group(2))

train_accuracy[0].append(iteration)

train_accuracy[1].append(trainacc)

train_loss_match = regex_train_output_loss.search(line)

if train_loss_match:

trainloss = float(train_loss_match.group(2))

train_loss[0].append(iteration)

train_loss[1].append(trainloss)

test_accuracy_match = regex_test_output_accuarcy.search(line)

if test_accuracy_match:

testacc = float(test_accuracy_match.group(2))

test_accuracy[0].append(iteration)

test_accuracy[1].append(testacc)

test_loss_match = regex_test_output_loss.search(line)

if test_loss_match:

testloss = float(test_loss_match.group(2))

test_loss[0].append(iteration)

test_loss[1].append(testloss)

return learning_rate, train_accuracy, train_loss, test_accuracy, test_loss

def caffe_plot(logfile):

learning_rate, train_accuracy, train_loss, test_accuracy, test_loss = parser_log(logfile)

plt.figure(1)

plt.title('loss vs iter')

plt.xlabel("iter")

plt.ylabel("loss")

plt.plot(test_loss[0], test_loss[1], 'b-', label='test loss')

plt.plot(train_loss[0], train_loss[1], 'r-', label='train loss')

plt.legend()

plt.show()

plt.savefig(r'D:\loss.png')

plt.figure(2)

plt.title("accuracy vs iter")

plt.xlabel("iter")

plt.ylabel("accuracy")

plt.plot(test_accuracy[0], test_accuracy[1], 'b-', label = 'test accuracy')

plt.plot(train_accuracy[0], train_accuracy[1], 'r-', label = 'train accuracy')

plt.legend()

plt.show()

plt.savefig(r'D:\accuracy.png')

plt.figure(3)

plt.title("lr vs iter")

plt.xlabel("iter")

plt.ylabel("learning rate")

plt.plot(learning_rate[0], learning_rate[1], 'b-', label = 'learning rate')

plt.legend()

plt.show()

plt.savefig(r'D:\learning_rate.png')

if __name__ == '__main__':

logfile = r"E:\DATA\FaceData\log\INFO2016-07-12T07-45-24_srcimg.txt"

caffe_plot(logfile)