"""Adapted from A3CTFPolicy to add V-trace. Keep in sync with changes to A3CTFPolicy and VtraceSurrogatePolicy.""" import numpy as np import logging import gym import ray from ray.rllib.agents.impala import vtrace_tf as vtrace from ray.rllib.models.tf.tf_action_dist import Categorical from ray.rllib.policy.sample_batch import SampleBatch from ray.rllib.policy.tf_policy_template import build_tf_policy from ray.rllib.policy.tf_policy import LearningRateSchedule, EntropyCoeffSchedule from ray.rllib.utils import force_list from ray.rllib.utils.framework import try_import_tf from ray.rllib.utils.tf_utils import explained_variance tf1, tf, tfv = try_import_tf() logger = logging.getLogger(__name__) class VTraceLoss: def __init__( self, actions, actions_logp, actions_entropy, dones, behaviour_action_logp, behaviour_logits, target_logits, discount, rewards, values, bootstrap_value, dist_class, model, valid_mask, config, vf_loss_coeff=0.5, entropy_coeff=0.01, clip_rho_threshold=1.0, clip_pg_rho_threshold=1.0, ): """Policy gradient loss with vtrace importance weighting. VTraceLoss takes tensors of shape [T, B, ...], where `B` is the batch_size. The reason we need to know `B` is for V-trace to properly handle episode cut boundaries. Args: actions: An int|float32 tensor of shape [T, B, ACTION_SPACE]. actions_logp: A float32 tensor of shape [T, B]. actions_entropy: A float32 tensor of shape [T, B]. dones: A bool tensor of shape [T, B]. behaviour_action_logp: Tensor of shape [T, B]. behaviour_logits: A list with length of ACTION_SPACE of float32 tensors of shapes [T, B, ACTION_SPACE[0]], ..., [T, B, ACTION_SPACE[-1]] target_logits: A list with length of ACTION_SPACE of float32 tensors of shapes [T, B, ACTION_SPACE[0]], ..., [T, B, ACTION_SPACE[-1]] discount: A float32 scalar. rewards: A float32 tensor of shape [T, B]. values: A float32 tensor of shape [T, B]. bootstrap_value: A float32 tensor of shape [B]. dist_class: action distribution class for logits. valid_mask: A bool tensor of valid RNN input elements (#2992). config: Trainer config dict. """ # Compute vtrace on the CPU for better perf. with tf.device("/cpu:0"): self.vtrace_returns = vtrace.multi_from_logits( behaviour_action_log_probs=behaviour_action_logp, behaviour_policy_logits=behaviour_logits, target_policy_logits=target_logits, actions=tf.unstack(actions, axis=2), discounts=tf.cast(~tf.cast(dones, tf.bool), tf.float32) * discount, rewards=rewards, values=values, bootstrap_value=bootstrap_value, dist_class=dist_class, model=model, clip_rho_threshold=tf.cast(clip_rho_threshold, tf.float32), clip_pg_rho_threshold=tf.cast(clip_pg_rho_threshold, tf.float32), ) self.value_targets = self.vtrace_returns.vs # The policy gradients loss. masked_pi_loss = tf.boolean_mask( actions_logp * self.vtrace_returns.pg_advantages, valid_mask ) self.pi_loss = -tf.reduce_sum(masked_pi_loss) self.mean_pi_loss = -tf.reduce_mean(masked_pi_loss) # The baseline loss. delta = tf.boolean_mask(values - self.vtrace_returns.vs, valid_mask) delta_squarred = tf.math.square(delta) self.vf_loss = 0.5 * tf.reduce_sum(delta_squarred) self.mean_vf_loss = 0.5 * tf.reduce_mean(delta_squarred) # The entropy loss. masked_entropy = tf.boolean_mask(actions_entropy, valid_mask) self.entropy = tf.reduce_sum(masked_entropy) self.mean_entropy = tf.reduce_mean(masked_entropy) # The summed weighted loss. self.total_loss = self.pi_loss - self.entropy * entropy_coeff # Optional vf loss (or in a separate term due to separate # optimizers/networks). self.loss_wo_vf = self.total_loss if not config["_separate_vf_optimizer"]: self.total_loss += self.vf_loss * vf_loss_coeff def _make_time_major(policy, seq_lens, tensor, drop_last=False): """Swaps batch and trajectory axis. Args: policy: Policy reference seq_lens: Sequence lengths if recurrent or None tensor: A tensor or list of tensors to reshape. drop_last: A bool indicating whether to drop the last trajectory item. Returns: res: A tensor with swapped axes or a list of tensors with swapped axes. """ if isinstance(tensor, list): return [_make_time_major(policy, seq_lens, t, drop_last) for t in tensor] if policy.is_recurrent(): B = tf.shape(seq_lens)[0] T = tf.shape(tensor)[0] // B else: # Important: chop the tensor into batches at known episode cut # boundaries. # TODO: (sven) this is kind of a hack and won't work for # batch_mode=complete_episodes. T = policy.config["rollout_fragment_length"] B = tf.shape(tensor)[0] // T rs = tf.reshape(tensor, tf.concat([[B, T], tf.shape(tensor)[1:]], axis=0)) # swap B and T axes res = tf.transpose(rs, [1, 0] + list(range(2, 1 + int(tf.shape(tensor).shape[0])))) if drop_last: return res[:-1] return res def build_vtrace_loss(policy, model, dist_class, train_batch): model_out, _ = model(train_batch) action_dist = dist_class(model_out, model) if isinstance(policy.action_space, gym.spaces.Discrete): is_multidiscrete = False output_hidden_shape = [policy.action_space.n] elif isinstance(policy.action_space, gym.spaces.MultiDiscrete): is_multidiscrete = True output_hidden_shape = policy.action_space.nvec.astype(np.int32) else: is_multidiscrete = False output_hidden_shape = 1 def make_time_major(*args, **kw): return _make_time_major( policy, train_batch.get(SampleBatch.SEQ_LENS), *args, **kw ) actions = train_batch[SampleBatch.ACTIONS] dones = train_batch[SampleBatch.DONES] rewards = train_batch[SampleBatch.REWARDS] behaviour_action_logp = train_batch[SampleBatch.ACTION_LOGP] behaviour_logits = train_batch[SampleBatch.ACTION_DIST_INPUTS] unpacked_behaviour_logits = tf.split(behaviour_logits, output_hidden_shape, axis=1) unpacked_outputs = tf.split(model_out, output_hidden_shape, axis=1) values = model.value_function() if policy.is_recurrent(): max_seq_len = tf.reduce_max(train_batch[SampleBatch.SEQ_LENS]) mask = tf.sequence_mask(train_batch[SampleBatch.SEQ_LENS], max_seq_len) mask = tf.reshape(mask, [-1]) else: mask = tf.ones_like(rewards) # Prepare actions for loss loss_actions = actions if is_multidiscrete else tf.expand_dims(actions, axis=1) # Inputs are reshaped from [B * T] => [(T|T-1), B] for V-trace calc. drop_last = policy.config["vtrace_drop_last_ts"] policy.loss = VTraceLoss( actions=make_time_major(loss_actions, drop_last=drop_last), actions_logp=make_time_major(action_dist.logp(actions), drop_last=drop_last), actions_entropy=make_time_major( action_dist.multi_entropy(), drop_last=drop_last ), dones=make_time_major(dones, drop_last=drop_last), behaviour_action_logp=make_time_major( behaviour_action_logp, drop_last=drop_last ), behaviour_logits=make_time_major( unpacked_behaviour_logits, drop_last=drop_last ), target_logits=make_time_major(unpacked_outputs, drop_last=drop_last), discount=policy.config["gamma"], rewards=make_time_major(rewards, drop_last=drop_last), values=make_time_major(values, drop_last=drop_last), bootstrap_value=make_time_major(values)[-1], dist_class=Categorical if is_multidiscrete else dist_class, model=model, valid_mask=make_time_major(mask, drop_last=drop_last), config=policy.config, vf_loss_coeff=policy.config["vf_loss_coeff"], entropy_coeff=policy.entropy_coeff, clip_rho_threshold=policy.config["vtrace_clip_rho_threshold"], clip_pg_rho_threshold=policy.config["vtrace_clip_pg_rho_threshold"], ) if policy.config.get("_separate_vf_optimizer"): return policy.loss.loss_wo_vf, policy.loss.vf_loss else: return policy.loss.total_loss def stats(policy, train_batch): drop_last = policy.config["vtrace"] and policy.config["vtrace_drop_last_ts"] values_batched = _make_time_major( policy, train_batch.get(SampleBatch.SEQ_LENS), policy.model.value_function(), drop_last=drop_last, ) return { "cur_lr": tf.cast(policy.cur_lr, tf.float64), "policy_loss": policy.loss.mean_pi_loss, "entropy": policy.loss.mean_entropy, "entropy_coeff": tf.cast(policy.entropy_coeff, tf.float64), "var_gnorm": tf.linalg.global_norm(policy.model.trainable_variables()), "vf_loss": policy.loss.mean_vf_loss, "vf_explained_var": explained_variance( tf.reshape(policy.loss.value_targets, [-1]), tf.reshape(values_batched, [-1]), ), } def grad_stats(policy, train_batch, grads): # We have support for more than one loss (list of lists of grads). if policy.config.get("_tf_policy_handles_more_than_one_loss"): grad_gnorm = [tf.linalg.global_norm(g) for g in grads] # Old case: We have a single list of grads (only one loss term and # optimizer). else: grad_gnorm = tf.linalg.global_norm(grads) return { "grad_gnorm": grad_gnorm, } def choose_optimizer(policy, config): if policy.config["opt_type"] == "adam": if policy.config["framework"] in ["tf2", "tfe"]: optim = tf.keras.optimizers.Adam(policy.cur_lr) if policy.config["_separate_vf_optimizer"]: return optim, tf.keras.optimizers.Adam(policy.config["_lr_vf"]) else: optim = tf1.train.AdamOptimizer(policy.cur_lr) if policy.config["_separate_vf_optimizer"]: return optim, tf1.train.AdamOptimizer(policy.config["_lr_vf"]) else: if policy.config["_separate_vf_optimizer"]: raise ValueError( "RMSProp optimizer not supported for separate" "vf- and policy losses yet! Set `opt_type=adam`" ) if tfv == 2: optim = tf.keras.optimizers.RMSprop( policy.cur_lr, config["decay"], config["momentum"], config["epsilon"] ) else: optim = tf1.train.RMSPropOptimizer( policy.cur_lr, config["decay"], config["momentum"], config["epsilon"] ) return optim def clip_gradients(policy, optimizer, loss): # Supporting more than one loss/optimizer. if policy.config["_tf_policy_handles_more_than_one_loss"]: optimizers = force_list(optimizer) losses = force_list(loss) assert len(optimizers) == len(losses) clipped_grads_and_vars = [] for optim, loss_ in zip(optimizers, losses): grads_and_vars = optim.compute_gradients( loss_, policy.model.trainable_variables() ) clipped_g_and_v = [] for g, v in grads_and_vars: if g is not None: clipped_g, _ = tf.clip_by_global_norm( [g], policy.config["grad_clip"] ) clipped_g_and_v.append((clipped_g[0], v)) clipped_grads_and_vars.append(clipped_g_and_v) policy.grads = [g for g_and_v in clipped_grads_and_vars for (g, v) in g_and_v] # Only one optimizer and and loss term. else: grads_and_vars = optimizer.compute_gradients( loss, policy.model.trainable_variables() ) grads = [g for (g, v) in grads_and_vars] policy.grads, _ = tf.clip_by_global_norm(grads, policy.config["grad_clip"]) clipped_grads_and_vars = list( zip(policy.grads, policy.model.trainable_variables()) ) return clipped_grads_and_vars def setup_mixins(policy, obs_space, action_space, config): LearningRateSchedule.__init__(policy, config["lr"], config["lr_schedule"]) EntropyCoeffSchedule.__init__( policy, config["entropy_coeff"], config["entropy_coeff_schedule"] ) VTraceTFPolicy = build_tf_policy( name="VTraceTFPolicy", get_default_config=lambda: ray.rllib.agents.impala.impala.DEFAULT_CONFIG, loss_fn=build_vtrace_loss, stats_fn=stats, grad_stats_fn=grad_stats, optimizer_fn=choose_optimizer, compute_gradients_fn=clip_gradients, before_loss_init=setup_mixins, mixins=[LearningRateSchedule, EntropyCoeffSchedule], get_batch_divisibility_req=lambda p: p.config["rollout_fragment_length"], )