loss_from_log.py

#!/usr/bin/env python
# Martin Kersner, 2016/01/13

from __future__ import print_function
import sys
import re
import numpy as np
import matplotlib.pyplot as plt

from utils import strstr

def main():
  log_files = process_arguments(sys.argv)

  train_iteration = []
  train_loss      = []
  test_iteration  = []
  test_loss       = []
  test_accuracy   = []

  pixel_accuracy        = []
  mean_accuracy         = []
  mean_IU               = []
  frequency_weighted_IU = []

  base_test_iter  = 0
  base_train_iter = 0

  for log_file in log_files:
    with open(log_file, 'rb') as f:
      if len(train_iteration) != 0:
        base_train_iter = train_iteration[-1]
        base_test_iter = test_iteration[-1]

      for line in f:
        # TRAIN NET
        if strstr(line, 'Iteration') and strstr(line, 'lr'):
          matched = match_iteration(line)
          train_iteration.append(int(matched.group(1))+base_train_iter)

        elif strstr(line, 'Train net output'):
          matched = match_loss(line)
          train_loss.append(float(matched.group(1)))

        elif strstr(line, 'pixel_accuracy'):
          matched = re.search(r'pixel_accuracy: (.*)', line)
          pixel_accuracy.append(float(matched.group(1)))

        elif strstr(line, 'mean_accuracy'):
          matched = re.search(r'mean_accuracy: (.*)', line)
          mean_accuracy.append(float(matched.group(1)))

        elif strstr(line, 'mean_IU'):
          matched = re.search(r'mean_IU: (.*)', line)
          mean_IU.append(float(matched.group(1)))

        elif strstr(line, 'frequency_weighted'):
          matched = re.search(r'frequency_weighted: (.*)', line)
          frequency_weighted_IU.append(float(matched.group(1)))


        # TEST NET
        elif strstr(line, 'Testing net'):
          matched = match_iteration(line)
          test_iteration.append(int(matched.group(1))+base_test_iter)

        elif strstr(line, 'Test net output'):
          matched = match_loss(line)
          if matched:
            test_loss.append(float(matched.group(1)))
          else:
            matched = match_accuracy(line)
            test_accuracy.append(float(matched.group(1)))

  print("TRAIN", train_iteration, train_loss)
  print("TEST", test_iteration, test_loss)
  print("ACCURACY", test_iteration, test_accuracy)

  # loss
  plt.plot(train_iteration, train_loss, 'k', label='Train loss')
  plt.plot(test_iteration, test_loss, 'r', label='Test loss')
  plt.legend()
  plt.ylabel('Loss')
  plt.xlabel('Number of iterations')
  plt.savefig('loss.png')

  # evaluation
  plt.clf()
  plt.plot(range(len(pixel_accuracy)), pixel_accuracy, 'k', label='pixel accuracy')
  plt.plot(range(len(mean_accuracy)), mean_accuracy, 'r', label='mean accuracy')
  plt.plot(range(len(mean_IU)), mean_IU, 'g', label='mean IU')
  plt.plot(range(len(frequency_weighted_IU)), frequency_weighted_IU, 'b', label='frequency weighted IU')
  plt.legend(loc=0)
  plt.savefig('evaluation.png')


def match_iteration(line):
  return re.search(r'Iteration (.*),', line)

def match_loss(line):
  return re.search(r'loss-ft = (.*) \(', line)

def match_accuracy(line):
  return re.search(r'seg-accuracy = (.*)', line)

def process_arguments(argv):
  if len(argv) < 2:
    help()

  log_files = argv[1:]
  return log_files

def help():
  print('Usage: python loss_from_log.py [LOG_FILE]+\n'
        'LOG_FILE is text file containing log produced by caffe.'
        'At least one LOG_FILE has to be specified.'
        'Files has to be given in correct order (the oldest logs as the first ones).'
        , file=sys.stderr)

  exit()

if __name__ == '__main__':
  main()