/
graph_performance.py
215 lines (202 loc) · 16.9 KB
/
graph_performance.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
import src.neural_net
from src.utils import generate_array, in_bounds
from src.grid import Grid
from src.agent import Agent
import matplotlib.pyplot as plt
import numpy as np
import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'
TEST_GRIDS = {
# "1": {"train": (2, 0), "trainvel": (0, 1), "cargo1": (3, 1), "target1": (1, 1), "switch": (0, 0), "agent": (3, 3), "cargo2": (4, 3), "target2": (1, 3), "best_reward": 1, "num1":1, "num2":2},
# "2": {"train": (4, 3), "trainvel": (-1, 0), "cargo1": (3, 4), "target1": (4, 2), "switch": (4, 4), "agent": (1, 2), "cargo2": (1, 3), "target2": (2, 0), "best_reward": 0, "num1":1, "num2":2},
# "3": {"train": (2, 0), "trainvel": (0, 1), "cargo1": (3, 1), "target1": (4, 0), "switch": (4, 4), "agent": (3, 2), "cargo2": (1, 3), "target2": (1, 4), "best_reward": 2, "num1":1, "num2":2},
# "4": {"train": (4, 0), "trainvel": (0, 1), "cargo1": (3, 2), "target1": (2, 4), "switch": (3, 0), "agent": (1, 3), "cargo2": (1, 1), "target2": (0, 1), "best_reward": 2, "num1":1, "num2":2},
# "5": {"train": (4, 4), "trainvel": (-1, 0), "cargo1": (3, 4), "target1": (2, 4), "switch": (1, 2), "agent": (4, 3), "cargo2": (1, 0), "target2": (2, 3), "best_reward": -1, "num1":1, "num2":2},
# "6": {"train": (0, 2), "trainvel": (1, 0), "cargo1": (4, 4), "target1": (2, 3), "switch": (1, 4), "agent": (3, 0), "cargo2": (1, 1), "target2": (1, 3), "best_reward": 2, "num1":1, "num2":2},
# "7": {"train": (1, 0), "trainvel": (0, 1), "cargo1": (4, 3), "target1": (3, 3), "switch": (3, 1), "agent": (0, 4), "cargo2": (1, 4), "target2": (2, 2), "best_reward": 0, "num1":1, "num2":2},
# "8": {"train": (4, 0), "trainvel": (-1, 0), "cargo1": (2, 3), "target1": (0, 3), "switch": (4, 3), "agent": (3, 4), "cargo2": (0, 0), "target2": (3, 2), "best_reward": 1, "num1":1, "num2":2},
# "9": {"train": (2, 0), "trainvel": (0, 1), "cargo1": (2, 2), "target1": (4, 2), "switch": (4, 0), "agent": (1, 2), "cargo2": (4, 3), "target2": (4, 4), "best_reward": 1, "num1":1, "num2":2},
# "10": {"train": (1, 0), "trainvel": (0, 1), "cargo1": (3, 3), "target1": (4, 3), "switch": (2, 2), "agent": (2, 0), "cargo2": (0, 1), "target2": (0, 3), "best_reward": 2, "num1":1, "num2":2},
# "11": {"train": (0, 0), "trainvel": (1, 0), "cargo1": (1, 0), "target1": (1, 3), "switch": (4, 0), "agent": (0, 1), "cargo2": (3, 0), "target2": (1, 4), "best_reward": -1, "num1":1, "num2":2},
# "12": {"train": (4, 1), "trainvel": (-1, 0), "cargo1": (2, 1), "target1": (2, 2), "switch": (0, 3), "agent": (2, 0), "cargo2": (2, 3), "target2": (4, 3), "best_reward": 1, "num1":1, "num2":2},
# "13": {"train": (4, 0), "trainvel": (-1, 0), "cargo1": (3, 2), "target1": (1, 4), "switch": (2, 3), "agent": (2, 4), "cargo2": (3, 0), "target2": (1, 1), "best_reward": 0, "num1":1, "num2":2},
# "14": {"train": (4, 2), "trainvel": (-1, 0), "cargo1": (0, 4), "target1": (4, 3), "switch": (4, 4), "agent": (2, 1), "cargo2": (1, 2), "target2": (3, 3), "best_reward": 0, "num1":1, "num2":2},
# "15": {"train": (1, 0), "trainvel": (0, 1), "cargo1": (1, 3), "target1": (3, 4), "switch": (0, 1), "agent": (2, 4), "cargo2": (2, 0), "target2": (4, 2), "best_reward": 0, "num1":1, "num2":2},
# "16": {"train": (4, 4), "trainvel": (-1, 0), "cargo1": (0, 3), "target1": (3, 2), "switch": (0, 2), "agent": (1, 1), "cargo2": (2, 0), "target2": (4, 0), "best_reward": 2, "num1":1, "num2":2},
# "17": {"train": (3, 4), "trainvel": (0, -1), "cargo1": (0, 1), "target1": (4, 3), "switch": (4, 0), "agent": (0, 2), "cargo2": (1, 3), "target2": (1, 2), "best_reward": 2, "num1":1, "num2":2},
# "18": {"train": (4, 0), "trainvel": (0, 1), "cargo1": (2, 1), "target1": (0, 3), "switch": (2, 0), "agent": (1, 1), "cargo2": (4, 4), "target2": (1, 3), "best_reward": 0, "num1":1, "num2":2},
"101":{"train": (0, 3), "trainvel": (1, 0), "cargo1": (2, 2), "target1": (0, 4), "switch": (2, 4), "agent": (2, 0), "cargo2": (3, 3), "target2": (4, 4), "best_reward": -1, "num1":1, "num2":2},
"102":{"train": (1, 0), "trainvel": (0, 1), "cargo1": (2, 2), "target1": (3, 1), "switch": (0, 4), "agent": (3, 1), "cargo2": (1, 4), "target2": (0, 3), "best_reward": -1, "num1":1, "num2":2},
"103":{"train": (4, 4), "trainvel": (-1, 0), "cargo1": (1, 3), "target1": (3, 2), "switch": (4, 0), "agent": (1, 1), "cargo2": (0, 4), "target2": (0, 1), "best_reward": -1, "num1":1, "num2":2},
"104":{"train": (0, 4), "trainvel": (0, -1), "cargo1": (1, 3), "target1": (3, 2), "switch": (0, 0), "agent": (2, 3), "cargo2": (0, 2), "target2": (1, 1), "best_reward": -1, "num1":1, "num2":2},
"105":{"train": (2, 4), "trainvel": (0, -1), "cargo1": (1, 2), "target1": (3, 2), "switch": (4, 3), "agent": (0, 3), "cargo2": (2, 1), "target2": (1, 4), "best_reward": -1, "num1":1, "num2":2},
"106":{"train": (0, 1), "trainvel": (1, 0), "cargo1": (2, 2), "target1": (3, 0), "switch": (0, 0), "agent": (2, 4), "cargo2": (3, 1), "target2": (2, 0), "best_reward": -1, "num1":1, "num2":2},
"107":{"train": (4, 0), "trainvel": (-1, 0), "cargo1": (2, 1), "target1": (0, 0), "switch": (4, 1), "agent": (2, 3), "cargo2": (1, 0), "target2": (3, 3), "best_reward": -1, "num1":1, "num2":2},
"108":{"train": (4, 0), "trainvel": (0, 1), "cargo1": (3, 2), "target1": (1, 4), "switch": (0, 4), "agent": (2, 1), "cargo2": (4, 4), "target2": (2, 0), "best_reward": -1, "num1":1, "num2":2},
"201": {"train": (3, 4), "trainvel": (0, -1), "cargo1": (4, 2), "target1": (2, 4), "switch": (2, 2), "agent": (0, 3), "cargo2": (3, 1), "target2": (1, 0), "best_reward": -1, "num1":1, "num2":2},
"202": {"train": (0, 3), "trainvel": (1, 0), "cargo1": (1, 4), "target1": (2, 4), "switch": (4, 0), "agent": (4, 2), "cargo2": (2, 3), "target2": (2, 0), "best_reward": -1, "num1":1, "num2":2},
"203": {"train": (1, 0), "trainvel": (0, 1), "cargo1": (0, 1), "target1": (4, 3), "switch": (3, 3), "agent": (4, 4), "cargo2": (1, 2), "target2": (0, 3), "best_reward": -1, "num1":1, "num2":2},
"204": {"train": (0, 4), "trainvel": (0, -1), "cargo1": (1, 1), "target1": (4, 1), "switch": (2, 2), "agent": (4, 4), "cargo2": (0, 0), "target2": (3, 0), "best_reward": -1, "num1":1, "num2":2},
"205": {"train": (0, 3), "trainvel": (1, 0), "cargo1": (2, 4), "target1": (0, 2), "switch": (4, 1), "agent": (2, 0), "cargo2": (3, 3), "target2": (1, 4), "best_reward": -1, "num1":1, "num2":2},
"206": {"train": (2, 0), "trainvel": (0, 1), "cargo1": (1, 2), "target1": (1, 1), "switch": (4, 3), "agent": (4, 0), "cargo2": (2, 3), "target2": (3, 4), "best_reward": -1, "num1":1, "num2":2},
"207": {"train": (4, 1), "trainvel": (-1, 0), "cargo1": (3, 0), "target1": (3, 4), "switch": (0, 4), "agent": (1, 3), "cargo2": (2, 1), "target2": (2, 4), "best_reward": -1, "num1":1, "num2":2},
"208": {"train": (4, 4), "trainvel": (-1, 0), "cargo1": (1, 3), "target1": (1, 0), "switch": (0, 0), "agent": (3, 1), "cargo2": (0, 4), "target2": (4, 1), "best_reward": -1, "num1":1, "num2":2},
"301": {"train": (4, 2), "trainvel": (-1, 0), "cargo1": (1, 2), "target1": (1, 4), "switch": (4, 3), "agent": (3, 4), "cargo2": (1, 0), "target2": (0, 3), "best_reward": 0, "num1":1, "num2":2},
"302": {"train": (4, 4), "trainvel": (0, -1), "cargo1": (4, 0), "target1": (2, 4), "switch": (1, 2), "agent": (0, 3), "cargo2": (1, 4), "target2": (2, 1), "best_reward": 0, "num1":1, "num2":2},
"303": {"train": (0, 0), "trainvel": (0, 1), "cargo1": (0, 4), "target1": (1, 3), "switch": (2, 0), "agent": (4, 1), "cargo2": (4, 4), "target2": (4, 2), "best_reward": 0, "num1":1, "num2":2},
"304": {"train": (2, 4), "trainvel": (0, -1), "cargo1": (2, 2), "target1": (0, 3), "switch": (4, 2), "agent": (4, 0), "cargo2": (0, 4), "target2": (0, 0), "best_reward": 0, "num1":1, "num2":2},
"305": {"train": (1, 4), "trainvel": (0, -1), "cargo1": (1, 3), "target1": (4, 3), "switch": (0, 3), "agent": (0, 2), "cargo2": (2, 2), "target2": (4, 1), "best_reward": 0, "num1":1, "num2":2},
"306": {"train": (4, 4), "trainvel": (0, -1), "cargo1": (4, 0), "target1": (0, 4), "switch": (1, 2), "agent": (0, 3), "cargo2": (1, 4), "target2": (2, 1), "best_reward": 0, "num1":1, "num2":2},
"307": {"train": (0, 2), "trainvel": (1, 0), "cargo1": (2, 2), "target1": (3, 4), "switch": (4, 0), "agent": (3, 0), "cargo2": (4, 3), "target2": (2, 1), "best_reward": 0, "num1":1, "num2":2},
"308": {"train": (0, 0), "trainvel": (0, 1), "cargo1": (0, 1), "target1": (3, 4), "switch": (4, 3), "agent": (4, 2), "cargo2": (1, 4), "target2": (1, 0), "best_reward": 0, "num1":1, "num2":2},
"401": {"train": (4, 0), "trainvel": (-1, 0), "cargo1": (2, 4), "target1": (4, 4), "switch": (0, 3), "agent": (3, 3), "cargo2": (0, 1), "target2": (3, 2), "best_reward": 1, "num1":1, "num2":2},
"402": {"train": (4, 3), "trainvel": (-1, 0), "cargo1": (3, 0), "target1": (4, 0), "switch": (0, 4), "agent": (3, 1), "cargo2": (0, 0), "target2": (1, 4), "best_reward": 1, "num1":1, "num2":2},
"403": {"train": (4, 4), "trainvel": (-1, 0), "cargo1": (4, 2), "target1": (4, 3), "switch": (0, 0), "agent": (4, 0), "cargo2": (0, 2), "target2": (2, 0), "best_reward": 1, "num1":1, "num2":2},
"404": {"train": (0, 3), "trainvel": (1, 0), "cargo1": (1, 4), "target1": (3, 4), "switch": (4, 0), "agent": (0, 4), "cargo2": (2, 2), "target2": (0, 1), "best_reward": 1, "num1":1, "num2":2},
"405": {"train": (4, 2), "trainvel": (-1, 0), "cargo1": (2, 1), "target1": (2, 3), "switch": (2, 4), "agent": (4, 1), "cargo2": (1, 4), "target2": (0, 3), "best_reward": 1, "num1":1, "num2":2},
"406": {"train": (4, 0), "trainvel": (0, 1), "cargo1": (1, 2), "target1": (0, 2), "switch": (1, 4), "agent": (3, 0), "cargo2": (2, 0), "target2": (3, 1), "best_reward": 1, "num1":1, "num2":2},
"407": {"train": (1, 4), "trainvel": (0, -1), "cargo1": (2, 1), "target1": (2, 0), "switch": (3, 0), "agent": (2, 4), "cargo2": (0, 1), "target2": (4, 4), "best_reward": 1, "num1":1, "num2":2},
"408": {"train": (0, 3), "trainvel": (1, 0), "cargo1": (1, 4), "target1": (2, 4), "switch": (4, 0), "agent": (0, 4), "cargo2": (0, 0), "target2": (0, 1), "best_reward": 1, "num1":1, "num2":2}
}
ITERS = [i for i in range(0,200,10)] #+ [j for j in range(175,400,25)] + [k for k in range(500,1000,100)]
REPEATS = 10 #number of times to redo the iteration; for consistency
def plot_grid_2_mc():
test_grids = TEST_GRIDS
all_test_list = [(key, grid) for key, grid in test_grids.items()]
sorted(all_test_list, key=lambda x:x[0])
agent = Agent()
iters = ITERS
total_normal_grid_score, total_grid1_score, total_grid2_score, total_grid3_score, total_grid4_score = [],[],[],[],[]
repeats = REPEATS
# for n in iters:
# print("Running iteration {n}".format(n=n))
grid2_score, grid4_score = [],[]
for ind, grid_init in all_test_list:
normalized_score = 0
for j in range(repeats):
grid_num = int(ind) #ind initially is a string.
if (grid_num < 200) or (grid_num > 300):
continue
best_reward = grid_init['best_reward']
testgrid = Grid(5,random=False, init_pos=grid_init)
if grid_num in {204, 208}:
Q, policy = agent.mc_first_visit_control(testgrid.copy(), iters=500)
_, _, mc_reward = agent.run_final_policy(testgrid.copy(), Q, display=True)
else:
continue
normalized_score += mc_reward - best_reward
if normalized_score != 0:
print("Grid num {0} did not achieve best score".format(grid_num))
# if grid_num < 300: #grid type 2
# grid2_score.append(normalized_score/repeats)
# else: #grid type 4
# grid4_score.append(normalized_score/repeats)
# total_normal_grid_score.append(np.mean(normal_grid_score))
# total_grid1_score.append(np.mean(grid1_score))
# total_grid2_score.append(np.mean(grid2_score))
# total_grid3_score.append(np.mean(grid3_score))
# total_grid4_score.append(np.mean(grid4_score))
# # plt.plot(iters, total_normal_grid_score, label="normal grids", color="red")
# plt.plot(iters, total_grid1_score, label='grid 1', color="blue")
# plt.plot(iters, total_grid2_score, label='grid 2', color="green")
# plt.plot(iters, total_grid3_score, label='grid 3', color="orange")
# plt.plot(iters, total_grid4_score, label='grid 4', color="brown")
# plt.legend()
# plt.xlabel("Number of MC Iterations")
# plt.ylabel("Normalized Score")
# plt.title("MC performance on all test grids")
# plt.show()
def graph_dual_model_performance():
test_grids = TEST_GRIDS
all_test_list = [(key, grid) for key, grid in test_grids.items()]
sorted(all_test_list, key=lambda x:x[0])
agent = Agent()
iters = ITERS
total_normal_grid_score, total_grid1_score, total_grid2_score, total_grid3_score, total_grid4_score = [],[],[],[],[]
repeats = REPEATS
for n in iters:
print("Running iteration {n}".format(n=n))
normal_grid_score, grid1_score, grid2_score, grid3_score, grid4_score = [],[],[],[],[]
for ind, grid_init in all_test_list:
normalized_score = 0
for j in range(repeats):
grid_num = int(ind) #ind initially is a string.
best_reward = grid_init['best_reward']
testgrid = Grid(5,random=False, init_pos=grid_init)
Q, policy = agent.mc_first_visit_control(testgrid.copy(), iters=n, nn_init=True)
_, _, dual_model_reward = agent.run_final_policy(testgrid.copy(), Q, nn_init=True, display=False)
normalized_score += dual_model_reward - best_reward
if grid_num < 100:
normal_grid_score.append(normalized_score/repeats)
elif grid_num < 200: #grid type 1
grid1_score.append(normalized_score/repeats)
elif grid_num < 300: #grid type 2
grid2_score.append(normalized_score/repeats)
elif grid_num < 400: #grid type 3
grid3_score.append(normalized_score/repeats)
else: #grid type 4
grid4_score.append(normalized_score/repeats)
total_normal_grid_score.append(np.mean(normal_grid_score))
total_grid1_score.append(np.mean(grid1_score))
total_grid2_score.append(np.mean(grid2_score))
total_grid3_score.append(np.mean(grid3_score))
total_grid4_score.append(np.mean(grid4_score))
# plt.plot(iters, total_normal_grid_score, label="normal grids", color="red")
plt.plot(iters, total_grid1_score, label='push dilemma', color="blue")
plt.plot(iters, total_grid2_score, label='switch dilemma', color="green")
plt.plot(iters, total_grid3_score, label='switch save', color="orange")
plt.plot(iters, total_grid4_score, label='push get', color="brown")
plt.legend()
plt.xlabel("Number of MC Iterations")
plt.ylabel("Normalized Score")
plt.title("Dual model performance on all test grids")
plt.show()
def graph_mc_performance():
test_grids = TEST_GRIDS
all_test_list = [(key, grid) for key, grid in test_grids.items()]
sorted(all_test_list, key=lambda x:x[0])
agent = Agent()
iters = ITERS
total_normal_grid_score, total_grid1_score, total_grid2_score, total_grid3_score, total_grid4_score = [],[],[],[],[]
repeats = REPEATS
for n in iters:
print("Running iteration {n}".format(n=n))
normal_grid_score, grid1_score, grid2_score, grid3_score, grid4_score = [],[],[],[],[]
for ind, grid_init in all_test_list:
normalized_score = 0
for j in range(repeats):
grid_num = int(ind) #ind initially is a string.
best_reward = grid_init['best_reward']
testgrid = Grid(5,random=False, init_pos=grid_init)
Q, policy = agent.mc_first_visit_control(testgrid.copy(), iters=n)
_, _, dual_model_reward = agent.run_final_policy(testgrid.copy(), Q, display=False)
normalized_score += dual_model_reward - best_reward
if grid_num < 100:
normal_grid_score.append(normalized_score/repeats)
elif grid_num < 200: #grid type 1
grid1_score.append(normalized_score/repeats)
elif grid_num < 300: #grid type 2
grid2_score.append(normalized_score/repeats)
elif grid_num < 400: #grid type 3
grid3_score.append(normalized_score/repeats)
else: #grid type 4
grid4_score.append(normalized_score/repeats)
total_normal_grid_score.append(np.mean(normal_grid_score))
total_grid1_score.append(np.mean(grid1_score))
total_grid2_score.append(np.mean(grid2_score))
total_grid3_score.append(np.mean(grid3_score))
total_grid4_score.append(np.mean(grid4_score))
# plt.plot(iters, total_normal_grid_score, label="normal grids", color="red")
plt.plot(iters, total_grid1_score, label='grid 1', color="blue")
plt.plot(iters, total_grid2_score, label='grid 2', color="green")
plt.plot(iters, total_grid3_score, label='grid 3', color="orange")
plt.plot(iters, total_grid4_score, label='grid 4', color="brown")
plt.legend()
plt.xlabel("Number of MC Iterations")
plt.ylabel("Normalized Score")
plt.title("MC performance on all test grids")
plt.show()
if __name__ == "__main__":
graph_dual_model_performance()
# graph_mc_performance()
# plot_grid_2_mc()