main.py

################################
#Required:
#python 2.7
#numpy
#matplotlib
#sklearn
#GPflow 0.3.5
#################################

from __future__ import division

import numpy as np
import matplotlib.pyplot as pl
import util
import svDGPOM

def load_parameters(case):
    #train_all, test, gfs, max_laser_distance, limits

    parameters = \
        {'sim': \
             ('data/icra17/icra17_sim_filled.csv',\
              49,\
              [-120, 120, -20, 120]),
        }

    return parameters[case]

def main():
      #load parameters
      fn_train, laser_max_distance, lims = load_parameters('sim')

      #load data
      #data is assumed to be already sampled from lidar beams as unoccupied and free
      #col0=scan index; col1=longitude; col2=latitude; col3=occupied/unoccupied
      X_all, Y_all = util.read_txy_csv(fn_train)
      X_plot = util.get_mesh_grid(resolution=0.5, limits=lims)

      #define the model
      model_svDGPOM = svDGPOM.svDGPOM(M=50, filter_informative_data_thresh=5) #M  = number of inducing points to kick it off with

      max_t = len(np.unique(X_all[:, 0]))
      for ith_scan in range(0, max_t, 2): #every other

         #extract data points of the ith scan
         ith_scan_indx = X_all[:, 0] == ith_scan
         print('\n{}th scan: size={}'.format(ith_scan, np.sum(ith_scan_indx)))

         Xtrain_ith_scan = X_all[ith_scan_indx, 1:]
         Ytrain_ith_scan = Y_all[ith_scan_indx, :]

         #let's train
         model_svDGPOM.learn(Xtrain_ith_scan, Ytrain_ith_scan)
         print('Size of X = {} data points'.format(model_svDGPOM.mdl.X.value.shape[0]))

         #predictions
         qry_svDGPOM = model_svDGPOM.predict(X_plot)
         qry_svDGPOM_mean = qry_svDGPOM[0] #mean prediction

         #plot
         pl.figure(figsize=(10,4))
         pl.subplot(121)
         pl.scatter(Xtrain_ith_scan[:,0][Ytrain_ith_scan[:,0]==1], Xtrain_ith_scan[:,1][Ytrain_ith_scan[:,0]==1], color='r', marker='.', alpha=0.5, label='Occupied')
         pl.scatter(Xtrain_ith_scan[:,0][Ytrain_ith_scan[:,0]==0], Xtrain_ith_scan[:,1][Ytrain_ith_scan[:,0]==0], color='b', marker='.', alpha=0.5, label='Free')
         pl.scatter(model_svDGPOM.mdl.Z.value[:, 0], model_svDGPOM.mdl.Z.value[:, 1], color='k', marker='x', alpha=0.7, label='Indicing')
         pl.legend()
         pl.title('Data | iter={}'.format(ith_scan))
         pl.subplot(122)
         pl.scatter(X_plot[:, 0], X_plot[:, 1], c=qry_svDGPOM_mean.ravel(), s=3, cmap='jet', vmin=0, vmax=1, edgecolor='');
         pl.colorbar()
         pl.title('VSDGPOM | iter={}'.format(ith_scan))
         #pl.show()
         pl.savefig('outputs/sim1_{}.png'.format(ith_scan))

if __name__ == '__main__':
    main()