/
arguments.py
143 lines (122 loc) · 9.28 KB
/
arguments.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
import argparse
import torch
import numpy as np
from tools import load_shape_dict, shotInfoPre, shapeProcessing
def get_args():
# Note that hyperparameters may originally be reported in ATARI game frames instead of agent steps
parser = argparse.ArgumentParser(description='Rainbow for IR BPP')
# Parameters for the reinforcement learning agent
parser.add_argument('--hidden-size', type=int, default=128, metavar='SIZE', help='Network hidden size')
parser.add_argument('--noisy-std', type=float, default=0.5, metavar='σ',
help='Initial standard deviation of noisy linear layers')
parser.add_argument('--atoms', type=int, default=31, metavar='C', help='Discretised size of value distribution')
parser.add_argument('--V-min', type=float, default=-1, metavar='V', help='Minimum of value distribution support')
parser.add_argument('--V-max', type=float, default=8, metavar='V', help='Maximum of value distribution support')
parser.add_argument('--target-update', type=int, default=int(1e3), metavar='τ',
help='Number of steps after which to update target network')
parser.add_argument('--multi-step', type=int, default=3, metavar='n', help='Number of steps for multi-step return')
parser.add_argument('--discount', type=float, default=0.99, metavar='γ', help='Discount factor')
parser.add_argument('--reward-clip', type=int, default=0, metavar='VALUE', help='Reward clipping (0 to disable)')
parser.add_argument('--learning-rate', type=float, default=0.0000625, metavar='η', help='Learning rate')
parser.add_argument('--batch-size', type=int, default=64, metavar='SIZE', help='Batch size')
parser.add_argument('--norm-clip', type=float, default=10, metavar='NORM', help='Max L2 norm for gradient clipping')
parser.add_argument('--memory-capacity', type=int, default=int(1e5), metavar='CAPACITY',
help='Experience replay memory capacity')
parser.add_argument('--replay-frequency', type=int, default=4, metavar='k',
help='Frequency of sampling from memory')
parser.add_argument('--priority-exponent', type=float, default=0.5, metavar='ω',
help='Prioritised experience replay exponent (originally denoted α)')
parser.add_argument('--priority-weight', type=float, default=1.0, metavar='β',
help='Initial prioritised experience replay importance sampling weight')
parser.add_argument('--id', type=str, default='default', help='Experiment ID')
parser.add_argument('--seed', type=int, default=123, help='Random seed')
parser.add_argument('--disable-cuda', action='store_true', help='Disable CUDA')
parser.add_argument('--T-max', type=int, default=int(50e6), metavar='STEPS',
help='Number of training steps (4x number of frames)')
parser.add_argument('--max-episode-length', type=int, default=int(108e3), metavar='LENGTH',
help='Max episode length in game frames (0 to disable)')
parser.add_argument('--history-length', type=int, default=1, metavar='T',
help='Number of consecutive states processed')
parser.add_argument('--architecture', type=str, default='canonical', choices=['canonical', 'dataset-efficient'],
metavar='ARCH', help='Network architecture')
parser.add_argument('--load-model', action='store_true', help='Load the trained model')
parser.add_argument('--learn-start', type=int, default=int(5e2), metavar='STEPS',
help='Number of steps before starting training')
parser.add_argument('--evaluation-interval', type=int, default=100000, metavar='STEPS',
help='Number of training steps between evaluations')
parser.add_argument('--evaluation-episodes-training', type=int, default=100, metavar='N',
help='Number of evaluation episodes to average over')
# TODO: Note that DeepMind's evaluation method is running the latest agent for 500K frames ever every 1M steps
parser.add_argument('--evaluation-size', type=int, default=500, metavar='N',
help='Number of transitions to use for validating Q')
parser.add_argument('--render', action='store_true', help='Display screen (testing only)')
parser.add_argument('--enable-cudnn', action='store_true', help='Enable cuDNN (faster but nondeterministic)')
parser.add_argument('--checkpoint-interval', default=10000,
help='How often to checkpoint the model, defaults to 0 (never checkpoint)')
parser.add_argument('--save-interval', default=1000, help='How often to save the model.')
parser.add_argument('--model-save-path',type=str, default='./logs/experiment', help='The path to save the trained model')
parser.add_argument('--disable-bzip-memory', action='store_true',
help='Don\'t zip the memory file. Not recommended (zipping is a bit slower and much, much smaller)')
parser.add_argument('--print-log-interval', type=int, default=10, help='How often to print training logs')
parser.add_argument('--adam-eps', type=float, default=1.5e-4, metavar='ε', help='Adam epsilon')
# Parameters for irregular shape packing
parser.add_argument('--envName', type=str, default='Physics-v0', help='The environment name for packing policy training and testing')
parser.add_argument('--dataset', type=str, default='blockout', help='The organized dataset folder for training') # blockout general kitchen abc
parser.add_argument('--device', type=int, default=0, help='GPU device ID')
parser.add_argument('--custom', type=str, default=None, help='Customized label for the experiment')
parser.add_argument('--hierachical', action='store_true', help='Use hierachical policy')
parser.add_argument('--bufferSize', type=int, default=1, help='Object buffer size') # 1 3 5 10
parser.add_argument('--num_processes', type=int, default=2, help='How many parallel processes used for training') # 16 1
parser.add_argument('--distributed', action='store_true', help='Use distributed training')
parser.add_argument('--samplePointsNum', type=int, default=1024, help='How many points to sample from the object surface')
parser.add_argument('--selectedAction', type=int, default=500, help='How many actions to select from the action space')
parser.add_argument('--maxBatch', type=int, default=2, help='How many batches for simulation')
parser.add_argument('--visual', action='store_true', help='Render the scene')
parser.add_argument('--resolutionA', type=float, default = 0.02, help='The resolution for the action space')
parser.add_argument('--resolutionH', type=float, default = 0.01, help='The resolution for the heightmap')
parser.add_argument('--resolutionZ', type=float, default = 0.01, help='The resolution for the z axis')
parser.add_argument('--resolutionRot', type=int, default = 8, help='The resolution for the rotation, 2 for cube, 4 for blockout, and 8 for the rest')
parser.add_argument('--locmodel', type=str, default=None, help='The path to load the trained location model to select location candidate.')
parser.add_argument('--ordmodel', type=str, default=None, help='(Optional) The path to load the trained order model to select object from the buffer')
parser.add_argument('--only-simulate-current', action='store_true', help='Only simulate the current item')
parser.add_argument('--non_blocking', action='store_true', help='Train actor and critic in non-blocking mode')
parser.add_argument('--time_limit', type=float, default = 0.01, help='Time limit for each simulation step when non_blocking is True')
parser.add_argument('--evaluate', action='store_true', help='Evaluate only')
parser.add_argument('--evaluation-episodes-test', type=int, default=2000, help='Number of evaluation episodes to average over')
args = parser.parse_args()
print('first hierachical',args.hierachical)
print(' ' * 26 + 'Options')
for k, v in vars(args).items():
print(' ' * 26 + k + ': ' + str(v))
args.objPath = './dataset/{}/shape_vhacd'.format(args.dataset)
args.pointCloud = './dataset/{}/pointCloud'.format(args.dataset)
args.dicPath = './dataset/{}/id2shape.pt'.format(args.dataset)
if args.dataset == 'kitchen':
args.dataSample = 'category'
else:
args.dataSample = 'instance'
args.categories = len(torch.load(args.dicPath))
args.bin_dimension = np.round([0.32, 0.32, 0.30], decimals=6)
args.ZRotNum = args.resolutionRot # Max: 4/8
args.objVecLen = 9
args.model = None
args.load_memory_path = None
args.save_memory_path = None
args.scale = [100, 100, 100] # fix it! don't change it!
args.meshScale = 1
args.heightResolution = args.resolutionZ
args.shapeDict, args.infoDict = load_shape_dict(args, True, scale=args.meshScale)
args.physics = True
args.heightMap = True
args.useHeightMap = True
args.globalView = True if args.evaluate else False
args.shotInfo = shotInfoPre(args, args.meshScale)
args.simulation = True
args.distributed = True
args.test_name = './dataset/{}/test_sequence.pt'.format(args.dataset)
args.shapeArray = shapeProcessing(args.shapeDict, args)
if args.evaluate:
args.num_processes = 1
# temp setting
# args.evaluate = True
return args