def setUp(self):
super(StreamingTensorNormalizerTest, self).setUp()
tf.compat.v1.reset_default_graph()
self._tensor_spec = tensor_spec.TensorSpec([3], tf.float32, 'obs')
self._tensor_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=self._tensor_spec)
self._dict_tensor_spec = {'a': self._tensor_spec, 'b': self._tensor_spec}
self._dict_tensor_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=self._dict_tensor_spec)
self.evaluate(tf.compat.v1.global_variables_initializer())
def testNormalizationFloat64(self):
spec = tensor_spec.TensorSpec([3], tf.float64, 'obs')
dict_tensor_spec = {'a': spec, 'b': spec}
dict_tensor_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=dict_tensor_spec)
self.evaluate(tf.compat.v1.global_variables_initializer())
as_tensor = functools.partial(tf.convert_to_tensor, dtype=tf.float64)
# Update with some initial values.
norm_obs = {
'a': np.random.randn(6, 2, 3),
'b': np.random.randn(6, 2, 3)
}
norm_obs_t = tf.nest.map_structure(as_tensor, norm_obs)
self.evaluate(dict_tensor_normalizer.update(norm_obs_t))
view_obs = {'a': np.random.randn(4, 3), 'b': np.random.randn(4, 3)}
view_obs_t = tf.nest.map_structure(as_tensor, view_obs)
observed = self.evaluate(
dict_tensor_normalizer.normalize(view_obs_t,
clip_value=-1,
variance_epsilon=1e-6))
norm_obs_avg = tf.nest.map_structure(lambda a: a.mean(axis=(0, 1)),
norm_obs)
norm_obs_std = tf.nest.map_structure(lambda a: a.std(axis=(0, 1)),
norm_obs)
expected = tf.nest.map_structure(
lambda obs, avg, std: (obs - avg) / std, view_obs, norm_obs_avg,
norm_obs_std)
self.assertAllClose(observed, expected)
def __init__(self,
optimizer=None,
mod_net=None,
observation_spec=None,
discount_factor=0.95,
num_epochs=15,
normalize_rewards=False,
reward_norm_clipping=10.0,
gradient_clipping=None,
name=None):
tf.Module.__init__(self, name=name)
self._optimizer = optimizer
self._mod_net = mod_net
self._discount_factor = discount_factor
self._num_epochs = num_epochs
self._reward_norm_clipping = reward_norm_clipping
self._gradient_clipping = gradient_clipping or 0.0
self._observation_spec = observation_spec
self._reward_normalizer = None
if normalize_rewards:
self._reward_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec.TensorSpec([], tf.float32),
scope='normalize_reward')
super(CatDQNTrainer, self).__init__()
def testGetVariables(self, dtype):
spec = tensor_spec.TensorSpec([3], dtype, 'obs')
t_normalizer = tensor_normalizer.StreamingTensorNormalizer(spec)
self.evaluate(tf.compat.v1.global_variables_initializer())
variables = t_normalizer.variables
for var in variables:
self.assertEqual(var.shape, spec.shape)
self.assertEqual(var.dtype, spec.dtype)
def testUpdate(self, dtype):
# Construct and evaluate normalized tensor. Should update mean &
# variance.
spec = tensor_spec.TensorSpec([3], dtype, 'obs')
dict_tensor_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec={
'a': spec,
'b': spec
})
self.evaluate(tf.compat.v1.global_variables_initializer())
np_array = np.array(
[[1.3, 4.2, 7.5], [8.3, 2.2, 9.5], [3.3, 5.2, 6.5]],
dtype.as_numpy_dtype)
tensors = {
'a': tf.constant(np_array, dtype=dtype),
'b': tf.constant(np_array, dtype=dtype)
}
def _compare(data):
n = data.shape[0]
expected_avg = data.mean(axis=0)
expected_var = data.var(axis=0)
expected_m2 = expected_var * n
expected_count = np.array([n] * 3)
new_count, new_avg, new_m2, _ = self.evaluate(
dict_tensor_normalizer.variables)
tf.nest.map_structure(
lambda v: self.assertAllClose(v, expected_count), new_count)
tf.nest.map_structure(
lambda v: self.assertAllClose(v, expected_avg), new_avg)
tf.nest.map_structure(
lambda v: self.assertAllClose(v, expected_m2), new_m2)
update_norm_vars = dict_tensor_normalizer.update(tensors)
self.evaluate(update_norm_vars)
_compare(data=np_array)
update_norm_vars = dict_tensor_normalizer.update(
tf.nest.map_structure(lambda t: t + 1.0, tensors))
self.evaluate(update_norm_vars)
_compare(data=np.concatenate((np_array, np_array + 1.0), axis=0))
update_norm_vars = dict_tensor_normalizer.update(
tf.nest.map_structure(lambda t: t - 1.0, tensors))
self.evaluate(update_norm_vars)
_compare(data=np.concatenate(
(np_array, np_array + 1.0, np_array - 1.0), axis=0))
def testMeanVariance(self, dtype):
t_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=tensor_spec.TensorSpec([3], dtype, 'obs'))
self.evaluate(tf.compat.v1.global_variables_initializer())
np_array = np.array(
[[1.3, 4.2, 7.5], [8.3, 2.2, 9.5], [3.3, 5.2, 6.5]],
dtype.as_numpy_dtype)
tensor = tf.constant(np_array, dtype=dtype)
self.evaluate(t_normalizer.update(tensor))
# Get new mean and variance, and make sure they changed.
new_mean, new_variance = self.evaluate(
t_normalizer._get_mean_var_estimates())
self.assertAllClose(np_array.mean(axis=0), new_mean[0])
self.assertAllClose(np_array.var(axis=0), new_variance[0])
def testFixedMean(self, dtype):
t_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=tensor_spec.TensorSpec([3], dtype, 'obs'))
self.evaluate(tf.compat.v1.global_variables_initializer())
np_array = np.array(
[[1.3, 4.2, 7.5], [8.3, 2.2, 9.5], [3.3, 5.2, 6.5]],
dtype.as_numpy_dtype)
tensor = tf.constant(np_array, dtype=dtype)
# It fails when run more than 41 iterations.
for i in range(41):
self.evaluate(t_normalizer.update(tensor))
# Get new mean and variance, and make sure they changed.
new_mean, _ = self.evaluate(t_normalizer._get_mean_var_estimates())
new_array = np.concatenate([np_array for _ in range(i + 1)],
axis=0)
self.assertAllClose(new_array.mean(axis=0), new_mean[0])
def testNormalizeVSNumpy(self, dtype):
t_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=tensor_spec.TensorSpec([3], dtype, 'obs'))
self.evaluate(tf.compat.v1.global_variables_initializer())
np_array = np.array(
[[1.3, 4.2, 7.5], [8.3, 2.2, 9.5], [3.3, 5.2, 6.5]],
dtype.as_numpy_dtype)
tensor = tf.constant(np_array, dtype=dtype)
self.evaluate(t_normalizer.update(tensor))
epsilon = 1e-6
# Get new mean and variance, and make sure they changed.
norm_obs = t_normalizer.normalize(tensor, variance_epsilon=epsilon)
exp_obs = ((np_array - np_array.mean(axis=0)) /
(np_array.std(axis=0) + epsilon))
self.assertAllClose(norm_obs, exp_obs)
def testReset(self, dtype):
# Get original mean and variance.
t_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec=tensor_spec.TensorSpec([3], dtype, 'obs'))
self.evaluate(tf.compat.v1.global_variables_initializer())
original_vars = self.evaluate(t_normalizer.variables)
# Construct and evaluate normalized tensor. Updates statistics.
np_array = np.array(
[[1.3, 4.2, 7.5], [8.3, 2.2, 9.5], [3.3, 5.2, 6.5]],
dtype.as_numpy_dtype)
tensor = tf.constant(np_array, dtype=dtype)
update_norm_vars = t_normalizer.update(tensor)
self.evaluate(update_norm_vars)
# Verify that the internal variables have been successfully reset.
self.evaluate(t_normalizer.reset())
reset_vars = self.evaluate(t_normalizer.variables)
self.assertAllClose(original_vars, reset_vars)
def __init__(self,
time_step_spec,
action_spec,
optimizer=None,
actor_net=None,
value_net=None,
importance_ratio_clipping=0.0,
lambda_value=0.95,
discount_factor=0.99,
entropy_regularization=0.0,
policy_l2_reg=0.0,
value_function_l2_reg=0.0,
value_pred_loss_coef=0.5,
num_epochs=25,
use_gae=False,
use_td_lambda_return=False,
normalize_rewards=True,
reward_norm_clipping=10.0,
normalize_observations=True,
log_prob_clipping=0.0,
kl_cutoff_factor=2.0,
kl_cutoff_coef=1000.0,
initial_adaptive_kl_beta=1.0,
adaptive_kl_target=0.01,
adaptive_kl_tolerance=0.3,
gradient_clipping=None,
check_numerics=False,
debug_summaries=False,
summarize_grads_and_vars=False,
train_step_counter=None,
name=None):
"""Creates a PPO Agent.
Args:
time_step_spec: A `TimeStep` spec of the expected time_steps.
action_spec: A nest of BoundedTensorSpec representing the actions.
optimizer: Optimizer to use for the agent.
actor_net: A function actor_net(observations, action_spec) that returns
tensor of action distribution params for each observation. Takes nested
observation and returns nested action.
value_net: A function value_net(time_steps) that returns value tensor from
neural net predictions for each observation. Takes nested observation
and returns batch of value_preds.
importance_ratio_clipping: Epsilon in clipped, surrogate PPO objective.
For more detail, see explanation at the top of the doc.
lambda_value: Lambda parameter for TD-lambda computation.
discount_factor: Discount factor for return computation.
entropy_regularization: Coefficient for entropy regularization loss term.
policy_l2_reg: Coefficient for l2 regularization of policy weights.
value_function_l2_reg: Coefficient for l2 regularization of value function
weights.
value_pred_loss_coef: Multiplier for value prediction loss to balance with
policy gradient loss.
num_epochs: Number of epochs for computing policy updates.
use_gae: If True (default False), uses generalized advantage estimation
for computing per-timestep advantage. Else, just subtracts value
predictions from empirical return.
use_td_lambda_return: If True (default False), uses td_lambda_return for
training value function. (td_lambda_return = gae_advantage +
value_predictions)
normalize_rewards: If true, keeps moving variance of rewards and
normalizes incoming rewards.
reward_norm_clipping: Value above and below to clip normalized reward.
normalize_observations: If true, keeps moving mean and variance of
observations and normalizes incoming observations.
log_prob_clipping: +/- value for clipping log probs to prevent inf / NaN
values. Default: no clipping.
kl_cutoff_factor: If policy KL changes more than this much for any single
timestep, adds a squared KL penalty to loss function.
kl_cutoff_coef: Loss coefficient for kl cutoff term.
initial_adaptive_kl_beta: Initial value for beta coefficient of adaptive
kl penalty.
adaptive_kl_target: Desired kl target for policy updates. If actual kl is
far from this target, adaptive_kl_beta will be updated.
adaptive_kl_tolerance: A tolerance for adaptive_kl_beta. Mean KL above (1
+ tol) * adaptive_kl_target, or below (1 - tol) * adaptive_kl_target,
will cause adaptive_kl_beta to be updated.
gradient_clipping: Norm length to clip gradients. Default: no clipping.
check_numerics: If true, adds tf.debugging.check_numerics to help find
NaN / Inf values. For debugging only.
debug_summaries: A bool to gather debug summaries.
summarize_grads_and_vars: If true, gradient summaries will be written.
train_step_counter: An optional counter to increment every time the train
op is run. Defaults to the global_step.
name: The name of this agent. All variables in this module will fall
under that name. Defaults to the class name.
Raises:
ValueError: If the actor_net is not a DistributionNetwork.
"""
if not isinstance(actor_net, network.DistributionNetwork):
raise ValueError(
'actor_net must be an instance of a DistributionNetwork.')
tf.Module.__init__(self, name=name)
self._optimizer = optimizer
self._actor_net = actor_net
self._value_net = value_net
self._importance_ratio_clipping = importance_ratio_clipping
self._lambda = lambda_value
self._discount_factor = discount_factor
self._entropy_regularization = entropy_regularization
self._policy_l2_reg = policy_l2_reg
self._value_function_l2_reg = value_function_l2_reg
self._value_pred_loss_coef = value_pred_loss_coef
self._num_epochs = num_epochs
self._use_gae = use_gae
self._use_td_lambda_return = use_td_lambda_return
self._reward_norm_clipping = reward_norm_clipping
self._log_prob_clipping = log_prob_clipping
self._kl_cutoff_factor = kl_cutoff_factor
self._kl_cutoff_coef = kl_cutoff_coef
self._adaptive_kl_target = adaptive_kl_target
self._adaptive_kl_tolerance = adaptive_kl_tolerance
self._gradient_clipping = gradient_clipping or 0.0
self._check_numerics = check_numerics
if initial_adaptive_kl_beta > 0.0:
# TODO(kbanoop): Rename create_variable.
self._adaptive_kl_beta = common.create_variable(
'adaptive_kl_beta', initial_adaptive_kl_beta, dtype=tf.float32)
else:
self._adaptive_kl_beta = None
self._reward_normalizer = None
if normalize_rewards:
self._reward_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec.TensorSpec([], tf.float32),
scope='normalize_reward')
self._observation_normalizer = None
if normalize_observations:
self._observation_normalizer = (
tensor_normalizer.StreamingTensorNormalizer(
time_step_spec.observation,
scope='normalize_observations'))
policy = greedy_policy.GreedyPolicy(
ppo_policy.PPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=False))
collect_policy = ppo_policy.PPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=True)
self._action_distribution_spec = (self._actor_net.output_spec)
super(PPOAgent,
self).__init__(time_step_spec,
action_spec,
policy,
collect_policy,
train_sequence_length=None,
debug_summaries=debug_summaries,
summarize_grads_and_vars=summarize_grads_and_vars,
train_step_counter=train_step_counter)
def __init__(
self,
time_step_spec,
action_spec,
optimizer=None,
actor_net=None,
value_net=None,
importance_ratio_clipping=0.0,
lambda_value=0.95,
discount_factor=0.99,
entropy_regularization=0.0,
policy_l2_reg=0.0,
value_function_l2_reg=0.0,
shared_vars_l2_reg=0.0,
value_pred_loss_coef=0.5,
num_epochs=25,
use_gae=False,
use_td_lambda_return=False,
normalize_rewards=True,
reward_norm_clipping=10.0,
normalize_observations=True,
log_prob_clipping=0.0,
kl_cutoff_factor=0.0,
kl_cutoff_coef=0.0,
initial_adaptive_kl_beta=0.0,
adaptive_kl_target=0.0,
adaptive_kl_tolerance=0.0,
gradient_clipping=None,
value_clipping=None,
check_numerics=False,
# TODO(b/150244758): Change the default to False once we move
# clients onto Reverb.
compute_value_and_advantage_in_train=True,
update_normalizers_in_train=True,
debug_summaries=False,
summarize_grads_and_vars=False,
train_step_counter=None,
name='AttentionPPOAgent'):
"""Creates a PPO Agent.
Args:
time_step_spec: A `TimeStep` spec of the expected time_steps.
action_spec: A nest of `BoundedTensorSpec` representing the actions.
optimizer: Optimizer to use for the agent, default to using
`tf.compat.v1.train.AdamOptimizer`.
actor_net: A `network.DistributionNetwork` which maps observations to
action distributions. Commonly, it is set to
`actor_distribution_network.ActorDistributionNetwork`.
value_net: A `Network` which returns the value prediction for input
states, with `call(observation, step_type, network_state)`. Commonly, it
is set to `value_network.ValueNetwork`.
importance_ratio_clipping: Epsilon in clipped, surrogate PPO objective.
For more detail, see explanation at the top of the doc.
lambda_value: Lambda parameter for TD-lambda computation.
discount_factor: Discount factor for return computation. Default to `0.99`
which is the value used for all environments from (Schulman, 2017).
entropy_regularization: Coefficient for entropy regularization loss term.
Default to `0.0` because no entropy bonus was used in (Schulman, 2017).
policy_l2_reg: Coefficient for L2 regularization of unshared actor_net
weights. Default to `0.0` because no L2 regularization was applied on
the policy network weights in (Schulman, 2017).
value_function_l2_reg: Coefficient for l2 regularization of unshared value
function weights. Default to `0.0` because no L2 regularization was
applied on the policy network weights in (Schulman, 2017).
shared_vars_l2_reg: Coefficient for l2 regularization of weights shared
between actor_net and value_net. Default to `0.0` because no L2
regularization was applied on the policy network or value network
weights in (Schulman, 2017).
value_pred_loss_coef: Multiplier for value prediction loss to balance with
policy gradient loss. Default to `0.5`, which was used for all
environments in the OpenAI baseline implementation. This parameters is
irrelevant unless you are sharing part of actor_net and value_net. In
that case, you would want to tune this coeeficient, whose value depends
on the network architecture of your choice.
num_epochs: Number of epochs for computing policy updates. (Schulman,2017)
sets this to 10 for Mujoco, 15 for Roboschool and 3 for Atari.
use_gae: If True (default False), uses generalized advantage estimation
for computing per-timestep advantage. Else, just subtracts value
predictions from empirical return.
use_td_lambda_return: If True (default False), uses td_lambda_return for
training value function; here: `td_lambda_return = gae_advantage +
value_predictions`. `use_gae` must be set to `True` as well to enable
TD -lambda returns. If `use_td_lambda_return` is set to True while
`use_gae` is False, the empirical return will be used and a warning
will be logged.
normalize_rewards: If true, keeps moving variance of rewards and
normalizes incoming rewards. While not mentioned directly in (Schulman,
2017), reward normalization was implemented in OpenAI baselines and
(Ilyas et al., 2018) pointed out that it largely improves performance.
You may refer to Figure 1 of https://arxiv.org/pdf/1811.02553.pdf for a
comparison with and without reward scaling.
reward_norm_clipping: Value above and below to clip normalized reward.
Additional optimization proposed in (Ilyas et al., 2018) set to `5` or
`10`.
normalize_observations: If `True`, keeps moving mean and variance of
observations and normalizes incoming observations. Additional
optimization proposed in (Ilyas et al., 2018). If true, and the
observation spec is not tf.float32 (such as Atari), please manually
convert the observation spec received from the environment to tf.float32
before creating the networks. Otherwise, the normalized input to the
network (float32) will have a different dtype as what the network
expects, resulting in a mismatch error.
Example usage: ```python observation_tensor_spec, action_spec,
time_step_tensor_spec = ( spec_utils.get_tensor_specs(env))
normalized_observation_tensor_spec = tf.nest.map_structure(
lambda s: tf.TensorSpec( dtype=tf.float32, shape=s.shape,
name=s.name ), observation_tensor_spec ) actor_net =
actor_distribution_network.ActorDistributionNetwork(
normalized_observation_tensor_spec, ...) value_net =
value_network.ValueNetwork( normalized_observation_tensor_spec,
...) # Note that the agent still uses the original
time_step_tensor_spec # from the environment. agent =
ppo_clip_agent.PPOClipAgent( time_step_tensor_spec, action_spec,
actor_net, value_net, ...) ```
log_prob_clipping: +/- value for clipping log probs to prevent inf / NaN
values. Default: no clipping.
kl_cutoff_factor: Only meaningful when `kl_cutoff_coef > 0.0`. A multipler
used for calculating the KL cutoff ( = `kl_cutoff_factor *
adaptive_kl_target`). If policy KL averaged across the batch changes
more than the cutoff, a squared cutoff loss would be added to the loss
function.
kl_cutoff_coef: kl_cutoff_coef and kl_cutoff_factor are additional params
if one wants to use a KL cutoff loss term in addition to the adaptive KL
loss term. Default to 0.0 to disable the KL cutoff loss term as this was
not used in the paper. kl_cutoff_coef is the coefficient to mulitply by
the KL cutoff loss term, before adding to the total loss function.
initial_adaptive_kl_beta: Initial value for beta coefficient of adaptive
KL penalty. This initial value is not important in practice because the
algorithm quickly adjusts to it. A common default is 1.0.
adaptive_kl_target: Desired KL target for policy updates. If actual KL is
far from this target, adaptive_kl_beta will be updated. You should tune
this for your environment. 0.01 was found to perform well for Mujoco.
adaptive_kl_tolerance: A tolerance for adaptive_kl_beta. Mean KL above `(1
+ tol) * adaptive_kl_target`, or below `(1 - tol) * adaptive_kl_target`,
will cause `adaptive_kl_beta` to be updated. `0.5` was chosen
heuristically in the paper, but the algorithm is not very sensitive to
it.
gradient_clipping: Norm length to clip gradients. Default: no clipping.
value_clipping: Difference between new and old value predictions are
clipped to this threshold. Value clipping could be helpful when training
very deep networks. Default: no clipping.
check_numerics: If true, adds `tf.debugging.check_numerics` to help find
NaN / Inf values. For debugging only.
compute_value_and_advantage_in_train: A bool to indicate where value
prediction and advantage calculation happen. If True, both happen in
agent.train(). If False, value prediction is computed during data
collection. This argument must be set to `False` if mini batch learning
is enabled.
update_normalizers_in_train: A bool to indicate whether normalizers are
updated as parts of the `train` method. Set to `False` if mini batch
learning is enabled, or if `train` is called on multiple iterations of
the same trajectories. In that case, you would need to use `PPOLearner`
(which updates all the normalizers outside of the agent). This ensures
that normalizers are updated in the same way as (Schulman, 2017).
debug_summaries: A bool to gather debug summaries.
summarize_grads_and_vars: If true, gradient summaries will be written.
train_step_counter: An optional counter to increment every time the train
op is run. Defaults to the global_step.
name: The name of this agent. All variables in this module will fall under
that name. Defaults to the class name.
Raises:
TypeError: if `actor_net` or `value_net` is not of type
`tf_agents.networks.Network`.
"""
if not isinstance(actor_net, network.Network):
raise TypeError(
'actor_net must be an instance of a network.Network.')
if not isinstance(value_net, network.Network):
raise TypeError(
'value_net must be an instance of a network.Network.')
# PPOPolicy validates these, so we skip validation here.
actor_net.create_variables(time_step_spec.observation)
value_net.create_variables(time_step_spec.observation)
tf.Module.__init__(self, name=name)
self._optimizer = optimizer
self._actor_net = actor_net
self._value_net = value_net
self._importance_ratio_clipping = importance_ratio_clipping
self._lambda = lambda_value
self._discount_factor = discount_factor
self._entropy_regularization = entropy_regularization
self._policy_l2_reg = policy_l2_reg
self._value_function_l2_reg = value_function_l2_reg
self._shared_vars_l2_reg = shared_vars_l2_reg
self._value_pred_loss_coef = value_pred_loss_coef
self._num_epochs = num_epochs
self._use_gae = use_gae
self._use_td_lambda_return = use_td_lambda_return
self._reward_norm_clipping = reward_norm_clipping
self._log_prob_clipping = log_prob_clipping
self._kl_cutoff_factor = kl_cutoff_factor
self._kl_cutoff_coef = kl_cutoff_coef
self._adaptive_kl_target = adaptive_kl_target
self._adaptive_kl_tolerance = adaptive_kl_tolerance
self._gradient_clipping = gradient_clipping or 0.0
self._value_clipping = value_clipping or 0.0
self._check_numerics = check_numerics
self._compute_value_and_advantage_in_train = (
compute_value_and_advantage_in_train)
self.update_normalizers_in_train = update_normalizers_in_train
if not isinstance(self._optimizer, tf.keras.optimizers.Optimizer):
logging.warning(
'Only tf.keras.optimizers.Optimizers are well supported, got a '
'non-TF2 optimizer: %s', self._optimizer)
self._initial_adaptive_kl_beta = initial_adaptive_kl_beta
if initial_adaptive_kl_beta > 0.0:
self._adaptive_kl_beta = common.create_variable(
'adaptive_kl_beta', initial_adaptive_kl_beta, dtype=tf.float32)
else:
self._adaptive_kl_beta = None
self._reward_normalizer = None
if normalize_rewards:
self._reward_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec.TensorSpec([], tf.float32),
scope='normalize_reward')
self._observation_normalizer = None
if normalize_observations:
self._observation_normalizer = (
tensor_normalizer.StreamingTensorNormalizer(
time_step_spec.observation,
scope='normalize_observations'))
self._advantage_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec.TensorSpec([], tf.float32),
scope='normalize_advantages')
policy = greedy_policy.GreedyPolicy(
attention_ppo_policy.AttentionPPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=False))
collect_policy = attention_ppo_policy.AttentionPPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=True,
compute_value_and_advantage_in_train=(
self._compute_value_and_advantage_in_train),
)
if isinstance(self._actor_net, network.DistributionNetwork):
# Legacy behavior
self._action_distribution_spec = self._actor_net.output_spec
else:
self._action_distribution_spec = self._actor_net.create_variables(
time_step_spec.observation)
# Set training_data_spec to collect_data_spec with augmented policy info,
# iff return and normalized advantage are saved in preprocess_sequence.
if self._compute_value_and_advantage_in_train:
training_data_spec = None
else:
training_policy_info = collect_policy.trajectory_spec.policy_info.copy(
)
training_policy_info.update({
'value_prediction':
collect_policy.trajectory_spec.policy_info['value_prediction'],
'return':
tensor_spec.TensorSpec(shape=[], dtype=tf.float32),
'advantage':
tensor_spec.TensorSpec(shape=[], dtype=tf.float32),
})
training_data_spec = collect_policy.trajectory_spec.replace(
policy_info=training_policy_info)
super(ppo_agent.PPOAgent,
self).__init__(time_step_spec,
action_spec,
policy,
collect_policy,
train_sequence_length=None,
training_data_spec=training_data_spec,
debug_summaries=debug_summaries,
summarize_grads_and_vars=summarize_grads_and_vars,
train_step_counter=train_step_counter)
# This must be built after super() which sets up self.data_context.
self._collected_as_transition = data_converter.AsTransition(
self.collect_data_context, squeeze_time_dim=False)
self._as_trajectory = data_converter.AsTrajectory(self.data_context,
sequence_length=None)
def __init__(self,
time_step_spec,
action_spec,
optimizer=None,
actor_net=None,
value_net=None,
importance_ratio_clipping=0.0,
lambda_value=0.95,
discount_factor=0.99,
entropy_regularization=0.0,
policy_l2_reg=0.0,
value_function_l2_reg=0.0,
shared_vars_l2_reg=0.0,
value_pred_loss_coef=0.5,
num_epochs=25,
use_gae=False,
use_td_lambda_return=False,
normalize_rewards=True,
reward_norm_clipping=10.0,
normalize_observations=True,
log_prob_clipping=0.0,
kl_cutoff_factor=2.0,
kl_cutoff_coef=1000.0,
initial_adaptive_kl_beta=1.0,
adaptive_kl_target=0.01,
adaptive_kl_tolerance=0.3,
gradient_clipping=None,
check_numerics=False,
compute_value_and_advantage_in_train=False,
debug_summaries=False,
summarize_grads_and_vars=False,
train_step_counter=None,
name=None):
"""Creates a PPO Agent.
Args:
time_step_spec: A `TimeStep` spec of the expected time_steps.
action_spec: A nest of `BoundedTensorSpec` representing the actions.
optimizer: Optimizer to use for the agent, default to using
`tf.compat.v1.train.AdamOptimizer`.
actor_net: A `network.DistributionNetwork` which maps observations to
action distributions. Commonly, it is set to
`actor_distribution_network.ActorDistributionNetwork`.
value_net: A `Network` which returns the value prediction for input
states, with `call(observation, step_type, network_state)`. Commonly, it
is set to `value_network.ValueNetwork`.
importance_ratio_clipping: Epsilon in clipped, surrogate PPO objective.
For more detail, see explanation at the top of the doc.
lambda_value: Lambda parameter for TD-lambda computation.
discount_factor: Discount factor for return computation. Default to `0.99`
which is the value used for all environments from (Schulman, 2017).
entropy_regularization: Coefficient for entropy regularization loss term.
Default to `0.0` because no entropy bonus was used in (Schulman, 2017).
policy_l2_reg: Coefficient for L2 regularization of unshared actor_net
weights. Default to `0.0` because no L2 regularization was applied on
the policy network weights in (Schulman, 2017).
value_function_l2_reg: Coefficient for l2 regularization of unshared value
function weights. Default to `0.0` because no L2 regularization was
applied on the policy network weights in (Schulman, 2017).
shared_vars_l2_reg: Coefficient for l2 regularization of weights shared
between actor_net and value_net. Default to `0.0` because no L2
regularization was applied on the policy network or value network
weights in (Schulman, 2017).
value_pred_loss_coef: Multiplier for value prediction loss to balance with
policy gradient loss. Default to `0.5`, which was used for all
environments in the OpenAI baseline implementation. This parameters is
irrelevant unless you are sharing part of actor_net and value_net. In
that case, you would want to tune this coeeficient, whose value depends
on the network architecture of your choice.
num_epochs: Number of epochs for computing policy updates. (Schulman,2017)
sets this to 10 for Mujoco, 15 for Roboschool and 3 for Atari.
use_gae: If True (default False), uses generalized advantage estimation
for computing per-timestep advantage. Else, just subtracts value
predictions from empirical return.
use_td_lambda_return: If True (default False), uses td_lambda_return for
training value function; here:
`td_lambda_return = gae_advantage + value_predictions`.
`use_gae` must be set to `True` as well to enable TD -lambda returns. If
`use_td_lambda_return` is set to True while `use_gae` is False, the
empirical return will be used and a warning will be logged.
normalize_rewards: If true, keeps moving variance of rewards and
normalizes incoming rewards. While not mentioned directly in (Schulman,
2017), reward normalization was implemented in OpenAI baselines and
(Ilyas et al., 2018) pointed out that it largely improves performance.
You may refer to Figure 1 of https://arxiv.org/pdf/1811.02553.pdf for a
comparison with and without reward scaling.
reward_norm_clipping: Value above and below to clip normalized reward.
Additional optimization proposed in (Ilyas et al., 2018) set to
`5` or `10`.
normalize_observations: If `True`, keeps moving mean and
variance of observations and normalizes incoming observations.
Additional optimization proposed in (Ilyas et al., 2018).
log_prob_clipping: +/- value for clipping log probs to prevent inf / NaN
values. Default: no clipping.
kl_cutoff_factor: Only meaningful when `kl_cutoff_coef > 0.0`. A multipler
used for calculating the KL cutoff ( =
`kl_cutoff_factor * adaptive_kl_target`). If policy KL averaged across
the batch changes more than the cutoff, a squared cutoff loss would
be added to the loss function.
kl_cutoff_coef: kl_cutoff_coef and kl_cutoff_factor are additional params
if one wants to use a KL cutoff loss term in addition to the adaptive KL
loss term. Default to 0.0 to disable the KL cutoff loss term as this was
not used in the paper. kl_cutoff_coef is the coefficient to mulitply by
the KL cutoff loss term, before adding to the total loss function.
initial_adaptive_kl_beta: Initial value for beta coefficient of adaptive
KL penalty. This initial value is not important in practice because the
algorithm quickly adjusts to it. A common default is 1.0.
adaptive_kl_target: Desired KL target for policy updates. If actual KL is
far from this target, adaptive_kl_beta will be updated. You should tune
this for your environment. 0.01 was found to perform well for Mujoco.
adaptive_kl_tolerance: A tolerance for adaptive_kl_beta. Mean KL above
`(1 + tol) * adaptive_kl_target`, or below
`(1 - tol) * adaptive_kl_target`,
will cause `adaptive_kl_beta` to be updated. `0.5` was chosen
heuristically in the paper, but the algorithm is not very
sensitive to it.
gradient_clipping: Norm length to clip gradients. Default: no clipping.
check_numerics: If true, adds `tf.debugging.check_numerics` to help find
NaN / Inf values. For debugging only.
compute_value_and_advantage_in_train: A bool to indicate where value
prediction and advantage calculation happen. If True, both happen in
agent.train(). If False, value prediction is computed during data
collection. This argument must be set to `False` if mini batch learning
is enabled.
debug_summaries: A bool to gather debug summaries.
summarize_grads_and_vars: If true, gradient summaries will be written.
train_step_counter: An optional counter to increment every time the train
op is run. Defaults to the global_step.
name: The name of this agent. All variables in this module will fall under
that name. Defaults to the class name.
Raises:
ValueError: If the actor_net is not a DistributionNetwork or value_net is
not a Network.
"""
if not isinstance(actor_net, network.DistributionNetwork):
raise ValueError(
'actor_net must be an instance of a network.DistributionNetwork.')
if not isinstance(value_net, network.Network):
raise ValueError('value_net must be an instance of a network.Network.')
actor_net.create_variables()
value_net.create_variables()
tf.Module.__init__(self, name=name)
self._optimizer = optimizer
self._actor_net = actor_net
self._value_net = value_net
self._importance_ratio_clipping = importance_ratio_clipping
self._lambda = lambda_value
self._discount_factor = discount_factor
self._entropy_regularization = entropy_regularization
self._policy_l2_reg = policy_l2_reg
self._value_function_l2_reg = value_function_l2_reg
self._shared_vars_l2_reg = shared_vars_l2_reg
self._value_pred_loss_coef = value_pred_loss_coef
self._num_epochs = num_epochs
self._use_gae = use_gae
self._use_td_lambda_return = use_td_lambda_return
self._reward_norm_clipping = reward_norm_clipping
self._log_prob_clipping = log_prob_clipping
self._kl_cutoff_factor = kl_cutoff_factor
self._kl_cutoff_coef = kl_cutoff_coef
self._adaptive_kl_target = adaptive_kl_target
self._adaptive_kl_tolerance = adaptive_kl_tolerance
self._gradient_clipping = gradient_clipping or 0.0
self._check_numerics = check_numerics
self._compute_value_and_advantage_in_train = (
compute_value_and_advantage_in_train)
if initial_adaptive_kl_beta > 0.0:
# TODO(kbanoop): Rename create_variable.
self._adaptive_kl_beta = common.create_variable(
'adaptive_kl_beta', initial_adaptive_kl_beta, dtype=tf.float32)
else:
self._adaptive_kl_beta = None
self._reward_normalizer = None
if normalize_rewards:
self._reward_normalizer = tensor_normalizer.StreamingTensorNormalizer(
tensor_spec.TensorSpec([], tf.float32), scope='normalize_reward')
self._observation_normalizer = None
if normalize_observations:
self._observation_normalizer = (
tensor_normalizer.StreamingTensorNormalizer(
time_step_spec.observation, scope='normalize_observations'))
policy = greedy_policy.GreedyPolicy(
ppo_policy.PPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=False))
collect_policy = ppo_policy.PPOPolicy(
time_step_spec=time_step_spec,
action_spec=action_spec,
actor_network=actor_net,
value_network=value_net,
observation_normalizer=self._observation_normalizer,
clip=False,
collect=True,
compute_value_and_advantage_in_train=(
self._compute_value_and_advantage_in_train),
)
self._action_distribution_spec = (self._actor_net.output_spec)
super(PPOAgent, self).__init__(
time_step_spec,
action_spec,
policy,
collect_policy,
train_sequence_length=None,
debug_summaries=debug_summaries,
summarize_grads_and_vars=summarize_grads_and_vars,
train_step_counter=train_step_counter)
