-
Notifications
You must be signed in to change notification settings - Fork 1
/
rollout.py
198 lines (159 loc) · 7.07 KB
/
rollout.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
"""
Reconstructing policies from checkpoints to create plots with rollouts.
"""
import numpy as np
import json
import matplotlib.pyplot as plt
import torch
from envs import CartpoleEnv, InvertedPendulumEnv, LinearizedInvertedPendulumEnv, PendubotEnv, VehicleLateralEnv, PowergridEnv
from models import ProjRENModel, ProjRNNModel, ProjRNNOldModel
from activations import LeakyReLU, Tanh
from deq_lib.solvers import broyden, anderson
from trainers import ProjectedPGTrainer, ProjectedPPOTrainer
env_map = {
"<class 'envs.inverted_pendulum.InvertedPendulumEnv'>": InvertedPendulumEnv,
"<class 'envs.linearized_inverted_pendulum.LinearizedInvertedPendulumEnv'>": LinearizedInvertedPendulumEnv,
"<class 'envs.cartpole.CartpoleEnv'>": CartpoleEnv,
"<class 'envs.pendubot.PendubotEnv'>": PendubotEnv,
"<class 'envs.vehicle.VehicleLateralEnv'>": VehicleLateralEnv,
"<class 'envs.powergrid.PowergridEnv'>": PowergridEnv,
"<class 'envs.learned_inverted_pendulum.LearnedInvertedPendulumEnv'>": InvertedPendulumEnv
}
model_map = {
"<class 'models.ProjREN.ProjRENModel'>": ProjRENModel,
"<class 'models.ProjRNN.ProjRNNModel'>": ProjRNNModel,
"<class 'models.ProjRNNOld.ProjRNNOldModel'>": ProjRNNOldModel,
}
phi_map = {
"<class 'activations.LeakyReLU'>": LeakyReLU,
"<class 'activations.Tanh'>": Tanh,
}
def load_agent(directory, checkpoint_path = None):
if checkpoint_path is None:
checkpoint_path = directory + '/checkpoint_001000/checkpoint-1000'
config_file = open(directory + '/params.json', 'r')
config = json.load(config_file)
config['env'] = env_map[config['env']]
config['model']['custom_model_config']['plant_cstor'] = config['env']
config['model']['custom_model'] = model_map[config['model']['custom_model']]
config['model']['custom_model_config']['phi_cstor'] = phi_map[config['model']['custom_model_config']['phi_cstor']]
if 'broyden' in config['model']['custom_model_config']['solver']:
config['model']['custom_model_config']['solver'] = broyden
else:
config['model']['custom_model_config']['solver'] = anderson
config['num_workers'] = 3
agent = ProjectedPPOTrainer(config = config)
agent.restore(checkpoint_path)
env = config['env'](config['env_config'])
return agent, env
def compute_rollout(agent, env, init_obs):
obs = init_obs
policy_state = agent.get_policy().get_initial_state()
prev_rew = 0
prev_act = np.zeros_like(env.action_space.sample())
rewards = []
actions = []
states = []
done = False
n_steps = 0
while not done:
states.append(env.state)
action, policy_state, _ = agent.compute_single_action(
obs, state = policy_state, prev_reward = prev_rew, prev_action = prev_act, explore = False
)
obs, rew, done, info = env.step(action, fail_on_state_space = False)
actions.append(action)
rewards.append(rew)
prev_rew = rew
n_steps += 1
if n_steps < env.time_max+1:
print('failure')
states = np.stack(states).T
actions = np.stack(actions).T
return states, actions
# Phase portraits
def phase_portrait(agent_dir, N_PER_DIM, ROLLOUT_LEN):
agent, env = load_agent(agent_dir, checkpoint_path=agent_dir + "/checkpoint_001667/checkpoint-1667")
rollouts = torch.zeros(N_PER_DIM, N_PER_DIM, ROLLOUT_LEN, env.state_size)
state_max = env.state_space.high/0.8
theta_points = torch.linspace(-state_max[0], state_max[0], N_PER_DIM)
thetadot_points = torch.linspace(-state_max[1], state_max[1], N_PER_DIM)
for i in range(N_PER_DIM):
for j in range(N_PER_DIM):
print(f'{i}, {j}')
init_obs = env.reset(np.array([theta_points[i], thetadot_points[j]]))
states, actions = compute_rollout(agent, env, init_obs)
rollouts[i, j] = torch.from_numpy(states.T)
return rollouts, env
N_PER_DIM = 7
ROLLOUT_LEN = 200 + 1
# true_agent_dir = '../ray_results/StableRwd_PPO/ProjRENModel_InvertedPendulumEnv_phiTanh_state2_hidden16_1_hidden_size=16_2022-03-25_15-29-16'
# learned_agent_dir = '../ray_results/StableRwd_PPO/ProjRENModel_LearnedInvertedPendulumEnv_phiTanh_state2_hidden16_0_2022-03-25_15-42-07'
# Learned B2 as well
learned_agent_dir = '../ray_results/StableRwd_PPO/ProjRENModel_LearnedInvertedPendulumEnv_phiTanh_state2_hidden16_0_2022-03-26_20-00-50'
true_agent_dir = '../ray_results/StableRwd_PPO/ProjRENModel_InvertedPendulumEnv_phiTanh_state2_hidden16_0_2022-03-26_20-02-35'
true_pp, _ = phase_portrait(true_agent_dir, N_PER_DIM, ROLLOUT_LEN)
learned_pp, env = phase_portrait(learned_agent_dir, N_PER_DIM, ROLLOUT_LEN)
plt.figure()
plt.subplot(121, title = 'True Plant Model', xlim=[-np.pi, np.pi], ylim=[-8, 8])
for i in range(N_PER_DIM):
for j in range(N_PER_DIM):
# print(f'{i}, {j}')
rollout = true_pp[i, j]
if torch.allclose(torch.zeros(env.state_size), rollout[-1]):
color = 'C2' # green
else:
color = 'C3' # red
plt.plot(rollout[:, 0], rollout[:, 1], color = color)
plt.xlabel('x1 (radians)')
plt.ylabel('x2 (radians/second)')
plt.subplot(122, title = 'Learned Plant Model', xlim=[-np.pi, np.pi], ylim=[-8, 8])
for i in range(N_PER_DIM):
for j in range(N_PER_DIM):
# print(f'{i}, {j}')
rollout = learned_pp[i, j]
if torch.allclose(torch.zeros(env.state_size), rollout[-1]):
color = 'C2' # green
else:
color = 'C3' # red
plt.plot(rollout[:, 0], rollout[:, 1], color = color)
plt.xlabel('x1 (radians)')
plt.show()
# Plotting rollouts of agents against each other from the same initial conditions.
# ren_dir = "../ray_results/Learned_InvPend/ProjRENModel_LearnedInvertedPendulumEnv_phiTanh_state2_hidden4_0_2022-03-25_10-08-38"
# rnn_dir = "../ray_results/Learned_InvPend/ProjRENModel_InvertedPendulumEnv_phiTanh_state2_hidden4_0_2022-03-25_10-08-52"
# ren_agent, _ = load_agent(ren_dir, checkpoint_path=ren_dir + "/checkpoint_000084/checkpoint-84")
# rnn_agent, env = load_agent(rnn_dir, checkpoint_path=rnn_dir + "/checkpoint_000084/checkpoint-84")#, checkpoint_path=rnn_dir + '/checkpoint_000010/checkpoint-10')
# # x vs t
# N_iters = 10
# ren_states = []
# ren_actions = []
# rnn_states = []
# rnn_actions = []
# for _ in range(N_iters):
# obs = env.reset()
# env_copy = copy.deepcopy(env)
# ren_state, ren_action = compute_rollout(ren_agent, env, obs)
# # assert ren_state.shape[1] == env.time_max + 1, 'fail long'
# ren_states.append(ren_state)
# ren_actions.append(ren_action)
# rnn_state, rnn_action = compute_rollout(rnn_agent, env_copy, obs)
# rnn_states.append(rnn_state)
# rnn_actions.append(rnn_action)
# plt.figure()
# plt.subplot(311)
# for i in range(N_iters):
# plt.plot(ren_states[i][0])
# # plt.plot(rnn_states[i][0])
# plt.title("theta")
# plt.subplot(312)
# for i in range(N_iters):
# plt.plot(ren_states[i][1])
# # plt.plot(rnn_states[i][1])
# plt.title("theta dot")
# plt.subplot(313)
# for i in range(N_iters):
# plt.plot(ren_actions[i][0]) #, 'tab:orange')
# # plt.plot(rnn_actions[i][0], 'tab:blue')
# plt.title("u")
# plt.show()