HsMM-FRvsFingerprintingSimulation.py

"""
1. Python code for the HsMM forward recursion algorithm
2. Corresponding paper: HsMM-based forward recursion algorithm for
real-time indoor localization
3. This is an illustrative example of how HMM works to localization a mobile phone user.
4. Author: Shuai Sun, 08/09/2020
5. Please report problems and bugs to sunshuai198903@163.com
"""

from p2mfile_parse import TxtManager
from hmm import HiddenSemiMarkov
import numpy as np
from plot_function import hmm_hsmm_comp
import matplotlib.pyplot as plt
import time

txt = TxtManager()
training_path = 'Data/_training'
trajectory_path = 'Data/_testing'
# extract the training and testing data, the RSS data is generated from a ray-tracing software
trained_data = txt.wirelessinsite_path(training_path, s_num=12, offset=9, ap_num=8)
trajectory_data = txt.wirelessinsite_path(trajectory_path, s_num=1, offset=1, ap_num=8)
time_length = trajectory_data[0].shape[0]

D = 50  # predefine the maximum possible duration time
# Note: BBD_para refers to paras for the state duration distribution, using a pre-trained beta-binomial distribution
# For the sake of illustration, we haven't included the corresponding code for HsMM model training using Baum-Welch
# The HsMM model training code will be released soon, together with a letter paper of using BBD for duration modelling.
# The trained state transition probability is also directly provided for now.
BBD_para = np.array([[.43, 5, D], [.35, 4.21, D], [.21, 2.25, D], [.46, 10, D], [.65, 8.88, D], [.53, 2.72, D],
                     [2, 3, D], [1, 40, D], [0.50, 5.53, D], [.65, 7.78, D], [.43, 6.49, D], [.43, 7.36, D]])

para_hsmm = {'state_num': 12, 'anchor_num': 8, 'initial_p': np.ones(12) / 12,
             'transition_p': np.array([[0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                                       [.68, 0, .32, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                                       [0, .29, 0, .3, 0, 0, .41, 0, 0, 0, 0, 0],
                                       [0, 0, .26, 0, .74, 0, 0, 0, 0, 0, 0, 0],
                                       [0, 0, 0, .71, 0, .29, 0, 0, 0, 0, 0, 0],
                                       [0, 0, 0, 0, 0.51, 0, 0, .49, 0, 0, 0, 0],
                                       [0, 0, .3, 0, 0, 0, 0, .31, .17, 0, 0, .21],
                                       [0, 0, 0, 0, 0, 0.48, .52, 0, 0, 0, 0, 0],
                                       [0, 0, 0, 0, 0, 0, .46, 0, 0, .54, 0, 0],
                                       [0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0],
                                       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
                                       [0, 0, 0, 0, 0, 0, .53, 0, 0, 0, .47, 0]]),
             'training_type': 'not_provided', 'likelihood_from': 'not_provided',
             'max_duration': D,
             'sojourn': {'type': ['BBD' for item in range(12)], 'parameter': BBD_para}}

hsmm = HiddenSemiMarkov(para_hsmm, specification=True)
mean_u, cov_u = hsmm.training(t_type='list', train_data=trained_data)
hsmm.set_gaussian(mean_u, cov_u)
# compute the likelihood (emission probability in the HsMM)
likelihood_value = hsmm.get_likelihood(trajectory_data[0])
hsmm.set_likelihood(likelihood_value)

start_time = time.time()
st_forward_hsmm = hsmm.forward_only_relax_scaling()[0]
end_time = time.time()
print("forward_hsmm running time:", (end_time - start_time) / time_length)

# -- To Evaluate the effect of scaling regarding practical implementation, use the following two commands
# alpha_nonscaled = hsmm.forward_only_relax_scaling()[2]
# alpha_scaled = hsmm.forward_process()[0]

st_max_likelihood = hsmm.maximum_likelihood(likelihood_value)

# This is the ground truth data, in each of the bracket is [state index, duration time]
st_ground = [[1, 8], [2, 18], [3, 23], [4, 36], [5, 46], [6, 62], [8, 78], [7, 120], [9, 129], [10, 138]]

hmm_hsmm_comp(
    [[st_max_likelihood, st_forward_hsmm]],
    [['Fingerprinting', 'HsMM-FR', 'Ground truth']], ground_truth=st_ground, fig_height=6, over_width=15, save_pdf=True)

plt.show()