def create_components(config, env):
action_dim = env.action_dim
encoder_config = config['encoder']
q_config = config['q']
if 'cnn_name' in encoder_config:
encoder_config['time_distributed'] = True
model = dict(
encoder=Encoder(encoder_config, name='encoder'),
q=Q(q_config, action_dim, name='q'),
target_encoder=Encoder(encoder_config, name='target_encoder'),
target_q=Q(q_config, action_dim, name='target_q'),
)
if config.get('rnn'):
rnn_config = config['rnn']
model.update({
'rnn': LSTM(rnn_config, name='rnn'),
'target_rnn': LSTM(rnn_config, name='target_rnn')
})
if config.get('actor'):
actor_config = config['actor']
model.update({
'actor':
Actor(actor_config, action_dim, name='actor'),
'target_actor':
Actor(actor_config, action_dim, name='target_actor')
})
return model
def create_components(config, env, **kwargs):
action_dim = env.action_dim
return dict(
encoder=Encoder(config['encoder'], name='encoder'),
q=Q(config['q'], action_dim, name='q'),
target_encoder=Encoder(config['encoder'], name='target_encoder'),
target_q=Q(config['q'], action_dim, name='target_q'),
)
def create_components(config, env, n_agents, **kwargs):
return dict(
q_encoder=Encoder(config['q_encoder'], name='encoder'),
q_rnn=rnn(config['q_rnn'], name='q_rnn'),
q=Q(config['q'], env.action_dim, name='q'),
qmixer=QMixer(config['q_mixer'], n_agents, name='qmix'),
target_q_encoder=Encoder(config['q_encoder'], name='target_encoder'),
target_q_rnn=rnn(config['q_rnn'], name='target_q_rnn'),
target_q=Q(config['q'], env.action_dim, name='target_q'),
target_qmixer=QMixer(config['q_mixer'], n_agents, name='target_qmix'),
)
def create_components(config, env, **kwargs):
action_dim = env.action_dim
return dict(
encoder=Encoder(config['encoder'], name='cnn'),
fpn=FractionProposalNetwork(config['fpn'], name='fpn'),
qe=QuantileEmbed(config['qe'], name='qe'),
q=Value(config['q'], action_dim, name='iqn'),
target_encoder=Encoder(config['encoder'], name='target_cnn'),
target_fpn=FractionProposalNetwork(config['fpn'], name='fpn'),
target_qe=QuantileEmbed(config['qe'], name='qe'),
target_q=Value(config['q'], action_dim, name='target_iqn'),
)
def create_components(config, env, **kwargs):
assert env.is_action_discrete
action_dim = env.action_dim
encoder_config = config['encoder']
quantile_config = config['quantile']
q_config = config['q']
return dict(
encoder=Encoder(encoder_config, name='encoder'),
quantile=Quantile(quantile_config, name='phi'),
q=Value(q_config, action_dim, name='q'),
target_encoder=Encoder(encoder_config, name='target_encoder'),
target_quantile=Quantile(quantile_config, name='target_phi'),
target_q=Value(q_config, action_dim, name='target_q'),
)
def create_components(config, env):
action_dim = env.action_dim
is_action_discrete = env.is_action_discrete
return dict(
encoder=Encoder(config['encoder']),
actor=Actor(config['actor'], action_dim, is_action_discrete),
value=Value(config['value'])
)
def create_components(config, env):
action_dim = env.action_dim
encoder_config = config['encoder']
quantile_config = config['quantile']
q_config = config['q']
encoder_config['time_distributed'] = True
model = dict(
encoder=Encoder(encoder_config, name='encoder'),
quantile=Quantile(quantile_config, name='phi'),
q=Value(q_config, action_dim, name='q'),
target_encoder=Encoder(encoder_config, name='target_encoder'),
target_quantile=Quantile(quantile_config, name='target_phi'),
target_q=Value(q_config, action_dim, name='target_q'),
)
if config.get('rnn'):
rnn_config = config['rnn']
model.update({
'rnn': LSTM(rnn_config, name='rnn'),
'target_rnn': LSTM(rnn_config, name='target_rnn')
})
return model