-
Notifications
You must be signed in to change notification settings - Fork 1
/
dt_main.py
164 lines (133 loc) · 6.06 KB
/
dt_main.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
import sys
import argparse
import time
import gym
import numpy as np
import torch
import matplotlib.pyplot as plt
import os
from utils import experience_replay, epsilon_scheduler, constants
from Agents import dt_agent, random_agent, dqn_agent
from utils.data_collection import DataCollector
from utils.data_transforms import image_transformation_just_norm, image_transformation_grayscale_crop_downscale, image_transformation_grayscale_crop_downscale_norm, image_transformation_grayscale_crop_downscale_norm_v2, image_transformation_grayscale_crop_downscale_v2
from torch.utils.data import Dataset, DataLoader
from utils.data_read import DataReader
config = dict()
agent = "dt"
env = None
def run():
# Global objects
global config
global agent
global env
dt_model = dt_agent.DTAgent(env, config)
if not config['train'] and config['load'] is not None:
# Evaluation mode
if config['dump_frequency'] is not None:
data_collector = DataCollector(config['output'], config['dump_frequency'])
else:
data_collector = None
# Load model
dt_model.load(config['load'])
# Evaluate
dt_model.model.eval()
evaluation_rewards = []
for eval_idx in range(config['eval_trajectories']):
episode_reward, episode_seq_len = dt_model.run_evaluation_traj(
data_transformation=image_transformation_grayscale_crop_downscale_norm_v2,
float_state=True,
data_collection_obj=data_collector,
debug_print_freq=config['print_frequency']
)
evaluation_rewards.append(episode_reward)
mean_eval_reward = np.mean(evaluation_rewards)
median_eval_reward = np.median(evaluation_rewards)
standard_deviation_eval_reward = np.std(evaluation_rewards)
print("Mean evaluation reward: ", mean_eval_reward)
print("Median evaluation reward: ", median_eval_reward)
print("Standard deviation evaluation reward: ", standard_deviation_eval_reward)
return
elif config['train'] and config['load'] is not None:
# Load and train a model
dt_model.load(config['load'])
# make the results folder if it doesn't exist
if not os.path.exists("results"):
os.makedirs("results")
# Train
print("Save frequency: ", config['model_save_frequency_dt'])
print("Evaluation frequency: ", config['evaluation_frequency_dt'])
print("Learning rate: ", config['learning_rate_dt'])
# TODO: trying out first k iters rather than last k
print("Loading data...")
reader = DataReader(
config['input_trajectory_path'],
store_transform=image_transformation_grayscale_crop_downscale_v2,
store_float_state=False,
return_transformation=image_transformation_just_norm,
return_float_state=True,
k_first_iters=200,
verbose_freq=50,
max_ep_load=config['data_trajectories'],
debug_print=True
)
print("Starting training...")
# Training mode
dt_model.train(
dataset=reader,
num_epochs=config['num_epochs'],
batch_size=1,
verbose=config['verbose'],
print_freq=config['print_frequency']
)
return
def main():
# Parse arguments
parser = argparse.ArgumentParser(
description='Playing MsPacman with Reinforcement Learning agents.')
parser.add_argument('-t', '--train', action='store_true', help='Train the agent')
parser.add_argument('-et', '--eval_trajectories', type=int, default=100, help='Number of trajectories to evaluate (in evaluation mode)')
parser.add_argument('-n', '--num_epochs', type=int, default=100000, help='Number of epochs to train')
parser.add_argument('-v', '--verbose', action='store_true', help='Verbose mode')
parser.add_argument('-pf', '--print_frequency', type=int, help='Frequency in episodes to print progress')
parser.add_argument('-df', '--dump_frequency', type=int, default=None, help="How many episodes between writing trajectories to outfile")
parser.add_argument('-i', '--input', type=str, default="random_trajectories.h5", help="Input trajectory path for data collection")
parser.add_argument('-o', '--eval_output', type=str, default="random_trajectories.h5", help="Output trajectory path for data collection")
parser.add_argument('--evaluation_frequency', type=int, help='Frequency in episodes to evaluate model')
parser.add_argument('-lr', '--learning_rate', type=float, help='Learning rate')
parser.add_argument('-l', '--load', type=str, help='Load model. Provide name of model file, with extension and folder')
parser.add_argument('-dt', '--data_trajectories', type=int, default=10000, help='Number of trajectories loaded from file.')
args = parser.parse_args()
# Set configuration
global config
config.update(constants.load()) # Load constants
# Set configuration based on arguments
config['agent'] = 'dt'
config['train'] = args.train
config['input_trajectory_path'] = args.input
config['num_epochs'] = args.num_epochs
config['verbose'] = args.verbose
config['load'] = args.load
config['dump_frequency'] = args.dump_frequency
config['output'] = args.eval_output
config['data_trajectories'] = args.data_trajectories
config['eval_trajectories'] = args.eval_trajectories
if args.print_frequency is not None:
config['print_frequency'] = args.print_frequency
if args.evaluation_frequency is not None:
config['evaluation_frequency'] = args.evaluation_frequency
if args.learning_rate is not None:
config['learning_rate_dt'] = args.learning_rate
print("Print frequency is: ", config['print_frequency'])
if config['train']:
config['save'] = True
else:
config['save'] = False
# Initialize environment
global env
env = gym.make('ALE/MsPacman-v5')
print('Playing MsPacman with {} agent'.format(config['agent']))
print('Mode: {}'.format('train' if args.train else 'evaluation'))
# Run
run()
if __name__ == '__main__':
main()