forked from sjyk/consumable-irl
/
Pinball.py
96 lines (77 loc) · 3 KB
/
Pinball.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
#!/usr/bin/env python
"""
Runs experiment with custom domain
"""
__author__ = "Richard Liaw"
import rlpy
from rlpy.Tools import deltaT, clock, hhmmss, getTimeStr
# from .. import visualize_trajectories as visual
import os
import yaml
import shutil
import inspect
import numpy as np
from rlpy.CustomDomains import RCIRL, Encoding, allMarkovReward
import domains
def load_yaml(file_path):
with open(file_path, 'r') as f:
ret_val = yaml.load(f)
return ret_val
def run_experiment_params(param_path='./params.yaml'):
params = type("Parameters", (), load_yaml(param_path))
def goalfn(state, goal, radius=0.1):
# Can be quickly modified to have a new radius per goal element
position = state[:2]
return (
np.linalg.norm(np.array(position)
- np.array(goal)) < radius
)
# # Load domain
def encode_trial():
rewards = list(params.domain_params['goalArray'])
encode = Encoding(rewards, goalfn)
return encode.strict_encoding
params.domain_params['goalfn'] = goalfn
params.domain_params['encodingFunction'] = encode_trial()
# params.domain_params['goalArray'] = params.domain_params['goalArray'][::4]
domain = eval(params.domain)(**params.domain_params)
# domain = eval(params.domain)()
#Load Representation
representation = eval(params.representation)(
domain,
**params.representation_params)
policy = eval(params.policy)(
representation,
**params.policy_params)
agent = eval(params.agent)(
policy,
representation,
discount_factor=domain.discount_factor,
**params.agent_params)
opt = {}
opt["exp_id"] = params.exp_id
opt["path"] = params.results_path + getTimeStr() + "/"
opt["max_steps"] = params.max_steps
# opt["max_eps"] = params.max_eps
opt["num_policy_checks"] = params.num_policy_checks
opt["checks_per_policy"] = params.checks_per_policy
opt["domain"] = domain
opt["agent"] = agent
if not os.path.exists(opt["path"]):
os.makedirs(opt["path"])
shutil.copy(param_path, opt["path"] + "params.yml")
shutil.copy(inspect.getfile(eval(params.domain)), opt["path"] + "domain.py")
shutil.copy(inspect.getfile(inspect.currentframe()), opt["path"] + "exper.py")
return eval(params.experiment)(**opt)
if __name__ == '__main__':
import sys
experiment = run_experiment_params(sys.argv[1])
import ipdb; ipdb.set_trace()
experiment.run(visualize_steps=0, # should each learning step be shown?
visualize_learning=False,
visualize_performance=False) # show policy / value function?
# saveTrajectories=False) # show performance runs?
# experiment.domain.showLearning(experiment.agent.representation)
# experiment.plotTrials(save=True)
# experiment.plot(save=True, x = "learning_episode") #, y="reward")
experiment.save()