exp4.py

   1 import numpy as np
   2 from Problems.synthetic_cosh import synthetic
   3 from analysis.analysis import error
   4 from graph.graph import Random
   5 from Optimizers import DOPTIMIZER as dopt
   6 from utilities import utilities as ut
   7 from utilities.plot_utils import plot_exp4
   8 import os
   9
  10
  11 ### creating times
  12 seed       = np.random.randint(12345)
  13 seed       = 8075
  14 np.random.seed(seed)
  15
  16 #### create asynchronous setup
  17 num_nodes  = 10
  18 dim        = 5
  19 comp_time_dist = 'random_uniform'
  20
  21 mincomp = np.array([1,1,1,1,1,1,1,1,1,1])
  22 maxcomp = np.array([5,10,15,20,25,30,35,40,45,50])
  23
  24 T_active_exp4, Tv_nodes_exp4, node_comp_time_exp4 = \
  25     ut.create_computation_time(num_nodes, max_iter=int(1e5), comp_time_dist=comp_time_dist, mean_comp=None,\
  26                                       min_comp=mincomp, max_comp=maxcomp, variance_comp=None, make_integer=True)
  27
  28
  29 #### some parameters of the algorithms
  30 learning_rate    = 0.01
  31 depoch    = 12000
  32 rho       = 0.01
  33 alpha     = 0.1
  34 gamma     = 0.5
  35 eta       = 1.0
  36 expScale  = 1/10.
  37
  38 drop_prob_0  = 0.0
  39 drop_prob_1  = 0.25
  40 drop_prob_2  = 0.5
  41 drop_prob_3  = 0.75
  42
  43
  44 #### Problem setup: parameters of the synthetic functions and constraints.
  45 prd = synthetic(seed, num_nodes, dim)
  46 error_prd = error(prd,np.zeros(num_nodes),0)
  47
  48
  49 #### Create gossip matrices
  50 zero_row_sum,zero_col_sum,row_stochastic,col_stochastic, N_out, neighbors  = Random(num_nodes, prob=0.8, Laplacian_dividing_factor= 2).directed()
  51
  52
  53 #### Run the optimization algorithms and compute the performance metrics
  54 np.random.seed(seed)
  55 x_asy_dagp1, _, _, _, _, Delay_mat_dagp_1 = \
  56     dopt.Asy_DAGP(T_active_exp4, Tv_nodes_exp4, prd, zero_row_sum, zero_col_sum, learning_rate, depoch, num_nodes, dim, rho, alpha, gamma, eta, neighbors, \
  57              cons = True, delay_type='exp', min_delay=None, max_delay=None, expScale_delay=expScale, \
  58                 drop_msg=True, drop_prob=drop_prob_1)
  59 np.random.seed(seed)
  60 x_asy_dagp2, _, _, _, _, Delay_mat_dagp_2 = \
  61     dopt.Asy_DAGP(T_active_exp4, Tv_nodes_exp4, prd, zero_row_sum, zero_col_sum, learning_rate, depoch, num_nodes, dim, rho, alpha, gamma, eta, neighbors, \
  62              cons = True, delay_type='exp', min_delay=None, max_delay=None, expScale_delay=expScale, \
  63                 drop_msg=True, drop_prob=drop_prob_2)
  64 np.random.seed(seed)
  65 x_asy_dagp3, _, _, _, _, Delay_mat_dagp_3 = \
  66     dopt.Asy_DAGP(T_active_exp4, Tv_nodes_exp4, prd, zero_row_sum, zero_col_sum, learning_rate, depoch, num_nodes, dim, rho, alpha, gamma, eta, neighbors, \
  67              cons = True, delay_type='exp', min_delay=None, max_delay=None, expScale_delay=expScale, \
  68                 drop_msg=True, drop_prob=drop_prob_3)
  69 np.random.seed(seed)
  70 x_asy_dagp0, _, _, _, _, Delay_mat_dagp_0 = \
  71     dopt.Asy_DAGP(T_active_exp4, Tv_nodes_exp4, prd, zero_row_sum, zero_col_sum, learning_rate, depoch, num_nodes, dim, rho, alpha, gamma, eta, neighbors, \
  72              cons = True, delay_type='exp', min_delay=None, max_delay=None, expScale_delay=expScale, \
  73                 drop_msg=True, drop_prob=drop_prob_0)
  74
  75
  76 f_asy_dagp0 = error_prd.cost_path(np.sum(x_asy_dagp0,  axis=1)/num_nodes)
  77 f_asy_dagp1 = error_prd.cost_path(np.sum(x_asy_dagp1,  axis=1)/num_nodes)
  78 f_asy_dagp2 = error_prd.cost_path(np.sum(x_asy_dagp2,  axis=1)/num_nodes)
  79 f_asy_dagp3 = error_prd.cost_path(np.sum(x_asy_dagp3,  axis=1)/num_nodes)
  80
  81
  82 #### save data and plot results
  83 plot_exp4(T_active_exp4, f_asy_dagp0, f_asy_dagp1, f_asy_dagp2, f_asy_dagp3, current_dir=os.path.dirname(os.path.abspath(__file__)), save_results_folder='exp4', plot_iter=depoch)
  84
  85