param

cfg/rad.cfg

@@ -33,6 +33,23 @@ dt = 0.01
 
 header = k, comm_radius, reward
 
+
+[4, 0.8]
+k = 4
+comm_radius = 4.0
+
+[3, 4.0]
+k = 3
+comm_radius = 4.0
+
+[2, 4.0]
+k = 2
+comm_radius = 4.0
+
+[1, 4.0]
+k = 1
+comm_radius = 4.0
+
 [4, 0.8]
 k = 4
 comm_radius = 0.8

results/dt.csv

@@ -1,4 +1,4 @@
-k, dt, reward
+k, dt, reward_mean, reward_std
 1, 0.1, -1270917.0522448036, 452924.35657368874
 1, 0.075, -30928.217654153723, 4563.656872878289
 1, 0.05, -14397.50841863872, 1673.3872385991804

results/plot_bargraph.py

@@ -0,0 +1,90 @@
+import matplotlib
+import matplotlib.pyplot as plt
+import csv
+import numpy as np
+from collections import OrderedDict
+
+font = {'family': 'serif',
+        'weight': 'bold',
+        'size': 14}
+matplotlib.rc('font', **font)
+
+_CENTRALIZED = 'Centr.'
+_DECENTRALIZED = 'Decentr.'
+
+def main():
+
+    # fig_fname = 'airsim_trained'
+    # fnames = ['airsim_trained.csv']
+
+    fig_fname = 'stoch_transfer_to_airsim'
+    fnames = ['stoch_transfer_to_airsim.csv']
+
+
+    # colors = {_CENTRALIZED: 'green', _DECENTRALIZED: 'red', '4': 'blue', '3': 'darkviolet', '2': 'orange', '1': 'gold'}
+    save_dir = 'fig/'
+
+    # fnames = ['rad.csv', 'rad_baseline.csv']
+    # xlabel = 'Comm. Radius'
+    k_ind = 0
+
+
+    mean_costs, std_costs = get_dict(fnames, k_ind)
+
+    # max_val, min_dec = get_max(mean_costs)
+    # max_val = max_val + 10.0
+    ylabel = 'Cost'
+    # title = ylabel + ' vs. ' + xlabel
+
+    # plot
+    fig, ax = plt.subplots()
+
+    ax.bar(mean_costs.keys(),  mean_costs.values(), yerr = std_costs.values())
+
+
+    # ax.legend()
+    plt.title('Trained on Stochastic Point-Mass Model')
+    # plt.ylim(top=max_val, bottom=0)
+    plt.xlabel('K')
+    plt.ylabel(ylabel)
+
+    plt.savefig(save_dir + fig_fname + '.eps', format='eps')
+    plt.show()
+
+
+def get_dict(fnames, k_ind):
+    mean_costs = OrderedDict()
+    std_costs = OrderedDict()
+
+    for fname in fnames:
+        with open(fname, 'r') as csvfile:
+            plots = csv.reader(csvfile, delimiter=',')
+            next(plots, None)
+            for row in plots:
+
+                if len(row) == 3:
+                    k = row[k_ind].strip()
+                    mean = float(row[1]) * -1.0
+                    std = float(row[2])
+
+                    mean_costs[k] = mean
+                    std_costs[k] = std
+
+    return mean_costs, std_costs
+
+
+def get_max(list_costs):
+    # compute average over diff seeds for each parameter combo
+    max_val = -1.0 * np.Inf
+    min_decentralized = 1.0 * np.Inf
+
+    for k in list_costs.keys():
+        for v in list_costs[k].keys():
+            if k != _DECENTRALIZED:
+                max_val = np.maximum(max_val, list_costs[k][v])
+            else:
+                min_decentralized = np.minimum(min_decentralized, list_costs[k][v])
+    return max_val, min_decentralized
+
+if __name__ == "__main__":
+    main()

results/stoch_transfer_to_airsim.csv

@@ -0,0 +1,5 @@
+k, reward, std
+1, -5.905066259211232, 2.2105622358605097
+2, -3.9234482654695833, 1.043596249170607
+3, -6.772500496052969, 2.5845733981345425
+4, -7.013095494140333, 2.6972839846893635