forked from justheuristic/mariewelt-mdp-med
-
Notifications
You must be signed in to change notification settings - Fork 0
/
RL_agents.py
154 lines (119 loc) · 6.16 KB
/
RL_agents.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
import numpy as np
import lasagne
import theano
from lasagne.layers import InputLayer, DimshuffleLayer, Pool2DLayer, DenseLayer, ExpressionLayer, batch_norm
from lasagne.regularization import regularize_network_params, l2
from super_agent import MdpAgent
from agentnet.environment import SessionPoolEnvironment
from agentnet.resolver import EpsilonGreedyResolver
from agentnet.memory import WindowAugmentation
from agentnet.learning import qlearning_n_step
from agentnet.agent import Agent
class BasicRLAgent(MdpAgent):
def __init__(self, pool, observation_shape, n_actions, n_parallel_games=1,
replay_seq_len=20, replay_batch_size=20, pool_size=None, n_steps=3, gamma=0.99):
"""
:type n_parallel_games: int
n_actions: int
"""
# Parameters for training
self.n_parallel_games = n_parallel_games
self.replay_seq_len = replay_seq_len
self.replay_batch_size = replay_batch_size
self.pool_size = pool_size
self.n_steps = n_steps
self.gamma = gamma
self.loss = None
# image observation
self.observation_layer = InputLayer(observation_shape)
self.n_actions = n_actions
self.resolver, self.agent = self.build_model()
weights = lasagne.layers.get_all_params(self.resolver, trainable=True)
self.applier_fun = self.agent.get_react_function()
# Prepare replay pool
env = SessionPoolEnvironment(observations=self.observation_layer,
actions=self.resolver,
agent_memories=self.agent.state_variables)
preceding_memory_states = list(pool.prev_memory_states)
# get interaction sessions
observation_tensor, action_tensor, reward_tensor, _, is_alive_tensor, _ = \
pool.interact(self.step, n_steps=self.replay_seq_len)
env.load_sessions(observation_tensor, action_tensor, reward_tensor, is_alive_tensor,
preceding_memory_states)
if pool_size is None:
batch_env = env
else:
batch_env = env.sample_session_batch(self.replay_batch_size)
self.loss = self.build_loss(batch_env)
self.eval_fun = self.build_eval_fun(batch_env)
updates = lasagne.updates.adadelta(self.loss, weights, learning_rate=0.01)
train_fun = theano.function([], [self.loss], updates=updates)
super(BasicRLAgent, self).__init__(env, pool, train_fun, pool_size, replay_seq_len)
def step(self, observation, prev_memories="zeros", batch_size=None):
"""
returns actions and new states given observation and prev state
Prev state in default setup should be [prev window,]
"""
# default to zeros
if batch_size is None:
batch_size = self.n_parallel_games
if prev_memories == 'zeros':
prev_memories = [np.zeros((batch_size,) + tuple(mem.output_shape[1:]),
dtype='float32')
for mem in self.agent.agent_states]
res = self.applier_fun(np.array(observation), *prev_memories)
action = res[0]
memories = res[1:]
return action, memories
def build_model(self):
# reshape to [batch, color, x, y] to allow for convolution layers to work correctly
observation_reshape = DimshuffleLayer(self.observation_layer, (0, 3, 1, 2))
observation_reshape = Pool2DLayer(observation_reshape, pool_size=(2, 2))
# memory
window_size = 5
# prev state input
prev_window = InputLayer((None, window_size) + tuple(observation_reshape.output_shape[1:]),
name="previous window state")
# our window
memory_layer = WindowAugmentation(observation_reshape,
prev_window,
name="new window state")
memory_dict = {memory_layer: prev_window}
# pixel-wise maximum over the temporal window (to avoid flickering)
memory_layer = ExpressionLayer(memory_layer, lambda a: a.max(axis=1),
output_shape=(None,) + memory_layer.output_shape[2:])
# neural network body
nn = batch_norm(lasagne.layers.Conv2DLayer(memory_layer, num_filters=16, filter_size=(8, 8), stride=(4, 4)))
nn = batch_norm(lasagne.layers.Conv2DLayer(nn, num_filters=32, filter_size=(4, 4), stride=(2, 2)))
nn = batch_norm(lasagne.layers.DenseLayer(nn, num_units=256))
# q_eval
policy_layer = DenseLayer(nn, num_units=self.n_actions, nonlinearity=lasagne.nonlinearities.linear,
name="QEvaluator")
# resolver
resolver = EpsilonGreedyResolver(policy_layer, name="resolver")
# all together
agent = Agent(self.observation_layer, memory_dict, policy_layer, resolver)
return resolver, agent
def build_loss(self, env):
_, _, _, _, qvalues_seq = self.agent.get_sessions(
env,
session_length=self.replay_seq_len,
batch_size=self.replay_batch_size,
optimize_experience_replay=True,
# unroll_scan=,
)
scaled_reward_seq = env.rewards
elwise_mse_loss = qlearning_n_step.get_elementwise_objective(qvalues_seq,
env.actions[0],
scaled_reward_seq,
env.is_alive,
n_steps=self.n_steps,
gamma_or_gammas=self.gamma, )
mse_loss = elwise_mse_loss.sum() / env.is_alive.sum()
reg_l2 = regularize_network_params(self.resolver, l2) * 10 ** -4
loss = mse_loss + reg_l2
return loss
def build_eval_fun(self, env):
mean_session_reward = env.rewards.sum(axis=1).mean() / self.replay_seq_len
eval_fun = theano.function([],[self.loss, mean_session_reward])
return eval_fun