the-last-thing/code/expt/bgt_cmp.py

#!/usr/bin/env python3

import sys
sys.path.insert(1, 'code/lib')
import argparse
from datetime import datetime
from geopy.distance import distance
import lmdk_bgt
import lmdk_lib
import numpy as np
from matplotlib import pyplot as plt
import time


def main(args):
  # The data files
  data_files = {
    'T-drive': '/home/manos/Cloud/Data/T-drive/Results.zip',
    'Geolife': '/home/manos/Cloud/Data/Geolife/Results.zip'
  }
  # Data related info
  data_info = {
    'T-drive': {
      'uid': 2,
      'lmdks': {
          0: {'dist': 0, 'per': 1000},   #   0.0%
         20: {'dist': 2095, 'per': 30},  #  19.6%
         40: {'dist': 2790, 'per': 30},  #  40.2%
         60: {'dist': 3590, 'per': 30},  #  59.9%
         80: {'dist': 4825, 'per': 30},  #  79.4%
        100: {'dist': 10350, 'per': 30}  # 100.0%
      }
    },
    'Geolife': {
      'uid': 97,
      'lmdks': {
          0: {'dist': 0, 'per': 100000},  #   0.0%
         20: {'dist': 205, 'per': 30},    #  19.8%
         40: {'dist': 450, 'per': 30},    #  41.7%
         60: {'dist': 725, 'per': 30},    #  59.2%
         80: {'dist': 855, 'per': 30},    #  82.1%
        100: {'dist': 50000, 'per': 30}   # 100.0%
      }
    }
  }
  # The data sets
  data_sets = {}
  # Load data sets
  for df in data_files:
    args.res = data_files[df]
    data_sets[df] = lmdk_lib.load_data(args, 'usrs_data')
  # Geo-I configuration
  # epsilon = level/radius
  # Radius is in meters
  bgt_conf = [
    {'epsilon': 1},
    # {'label': 'ln(2)/200', 'epsilon': 0.0035, 'level': 0.69314718056, 'radius': 200},
    # {'label': 'ln(4)/200', 'epsilon': 0.0069, 'level': 1.38629436112, 'radius': 200},
    # {'label': 'ln(6)/200', 'epsilon': 0.0090, 'level': 1.79175946923, 'radius': 200}
  ]

  # Number of methods
  n = 6
  # Width of bars
  bar_width = 1/(n + 1)
  # The x axis
  x_i = np.arange(len(list(data_info.values())[0]['lmdks']))
  x_margin = bar_width*(n/2 + 1)

  for d in data_sets:
    # d = 'T-drive'
    # d = 'Geolife'
    print('\n##############################', d, '\n')
    args.res = data_files[d]
    data = data_sets[d]
    # Truncate trajectory according to arguments
    seq = data[data[:,0]==data_info[d]['uid'], :][:args.time]

    # Initialize plot
    lmdk_lib.plot_init()
    # The x axis
    plt.xticks(x_i, np.array([key for key in data_info[d]['lmdks']]).astype(int))
    plt.xlabel('Landmarks percentage')  # Set x axis label.
    plt.xlim(x_i.min() - x_margin, x_i.max() + x_margin)
    # The y axis
    plt.ylabel('Mean absolute error (m)')  # Set y axis label.
    plt.yscale('log')
    plt.ylim(1, 100000)
    # Bar offset
    x_offset = -(bar_width/2)*(n - 1)

    mae_u = np.zeros(len(data_info[d]['lmdks']))
    mae_s = np.zeros(len(data_info[d]['lmdks']))
    mae_a = np.zeros(len(data_info[d]['lmdks']))
    mae_r = np.zeros(len(data_info[d]['lmdks']))
    mae_d = np.zeros(len(data_info[d]['lmdks']))
    mae_i = np.zeros(len(data_info[d]['lmdks']))
    for i, lmdk in enumerate(data_info[d]['lmdks']):
      # Find landmarks
      args.dist = data_info[d]['lmdks'][lmdk]['dist']
      args.per = data_info[d]['lmdks'][lmdk]['per']
      lmdks = lmdk_lib.find_lmdks(seq, args)[:args.time]
      # Print stats
      lmdk_lib.lmdks_stats(args, lmdks)
      # # Find long enough sequences
      # usrs = np.unique(data[:,0])
      # for usr_i, usr in enumerate(usrs):
      #   traj = data[data[:,0]==usr, :]
      #   if(len(traj)) >= 1000 and len(traj) < 2000:
      #     print(usr, len(traj))
      for bgt in bgt_conf:
        for _ in range(args.iter):
          # Skip
          rls_data_s, _ = lmdk_bgt.skip(seq, lmdks, bgt['epsilon'])
          mae_s[i] += lmdk_bgt.mae(seq, rls_data_s)/args.iter

          # Uniform
          rls_data_u, _ = lmdk_bgt.uniform_r(seq, lmdks, bgt['epsilon'])
          mae_u[i] += lmdk_bgt.mae(seq, rls_data_u)/args.iter

          # Adaptive
          rls_data_a, _, _ = lmdk_bgt.adaptive(seq, lmdks, bgt['epsilon'], .5, .5)
          mae_a[i] += lmdk_bgt.mae(seq, rls_data_a)/args.iter

          # Sample
          rls_data_r, _, _ = lmdk_bgt.sample(seq, lmdks, bgt['epsilon'])
          mae_r[i] += lmdk_bgt.mae(seq, rls_data_r)/args.iter

          # Discount
          rls_data_d, _, _ = lmdk_bgt.discount(seq, lmdks, bgt['epsilon'])
          mae_d[i] += lmdk_bgt.mae(seq, rls_data_d)/args.iter

          # Incremental
          rls_data_i, _, _ = lmdk_bgt.incremental(seq, lmdks, bgt['epsilon'], .5)
          mae_i[i] += lmdk_bgt.mae(seq, rls_data_i)/args.iter

        # print(
        #   '\nEpsilon  : %f\n'
        #   'Sampled  : %d%% (%d/%d)\n'
        #   'Landmarks: %d%% (%d/%d)\n'
        #   %(bgt['epsilon'], 100*(len(seq) - skipped)/len(seq), len(seq) - skipped, len(seq), 100*len(lmdks)/len(seq), len(lmdks), len(seq))
        # )
        # s, l = lmdk_lib.simplify_data(seq, lmdks)
        # # Validate the process
        # lmdk_bgt.validate_bgts(s, l, bgt['epsilon'], bgts)

        # # Analysis
        # lmdk_bgt.utility_analysis(seq, lmdks, rls_data, bgt['epsilon'])

    plt.bar(
      x_i + x_offset,
      mae_s,
      bar_width,
      label='Skip',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width
    # Plot bars
    plt.bar(
      x_i + x_offset,
      mae_u,
      bar_width,
      label='Uniform',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width
    plt.bar(
      x_i + x_offset,
      mae_a,
      bar_width,
      label='Adaptive',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width
    plt.bar(
      x_i + x_offset,
      mae_r,
      bar_width,
      label='Sample',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width
    plt.bar(
      x_i + x_offset,
      mae_d,
      bar_width,
      label='Discount',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width
    plt.bar(
      x_i + x_offset,
      mae_i,
      bar_width,
      label='Incremental',
      linewidth=lmdk_lib.line_width
    )
    x_offset += bar_width

    path = str('rslt/bgt_cmp/' + d)
    # Plot legend
    lmdk_lib.plot_legend()
    # Show plot
    # plt.show()
    # Save plot
    lmdk_lib.save_plot(path + '.pdf')
    print('[OK]', flush=True)




def parse_args():
  '''
    Parse arguments.

    Optional:
      dist - The coordinates distance threshold in meters.
      per  - The timestaps period threshold in mimutes.
      time - The total timestamps.
      iter - The total iterations.
  '''
  # Create argument parser.
  parser = argparse.ArgumentParser()

  # Mandatory arguments.

  # Optional arguments.
  parser.add_argument('-l', '--dist', help='The coordinates distance threshold in meters.', type=int, default=200)
  parser.add_argument('-p', '--per', help='The timestaps period threshold in mimutes.', type=int, default=30)
  parser.add_argument('-r', '--res', help='The results archive file.', type=str, default='/home/manos/Cloud/Data/T-drive/Results.zip')
  parser.add_argument('-t', '--time', help='The total timestamps.', type=int, default=1000)
  parser.add_argument('-i', '--iter', help='The total iterations.', type=int, default=1)

  # Parse arguments.
  args = parser.parse_args()

  return args


if __name__ == '__main__':
  try:
    start_time = time.time()
    main(parse_args())
    end_time = time.time()
    print('##############################')
    print('Time   : %.4fs' % (end_time - start_time))
    print('##############################')
  except KeyboardInterrupt:
    print('Interrupted by user.')
    exit()