Fix codespell issue

Author: emailweixu

alf/algorithms/agent.py

@@ -506,7 +506,7 @@ def preprocess_experience(self, root_inputs, rollout_info, batch_info):
 
     def summarize_rollout(self, experience):
         """First call ``RLAlgorithm.summarize_rollout()`` to summarize basic
-        rollout statisics. If the rl algorithm has overridden this function,
+        rollout statistics. If the rl algorithm has overridden this function,
         then also call its customized version.
         """
         super(Agent, self).summarize_rollout(experience)

alf/algorithms/mcts_algorithm.py

@@ -227,6 +227,7 @@ def _add_node(name: str, properties: dict):
 
 @alf.configurable
 class MCTSAlgorithm(OffPolicyAlgorithm):
+    # codespell:ignore-begin
     r"""Monte-Carlo Tree Search algorithm.
 
     The code largely follows the pseudocode of
@@ -300,6 +301,8 @@ class MCTSAlgorithm(OffPolicyAlgorithm):
       extend these k' paths are most promising according to the UCB scores.
     """
 
+    # codespell:ignore-end
+
     def __init__(
         self,
         observation_spec,

alf/algorithms/muzero_representation_learner.py

@@ -60,6 +60,7 @@
 
 @alf.configurable
 class MuzeroRepresentationImpl(OffPolicyAlgorithm):
+    # codespell:ignore-begin
     """MuZero-style Representation Learner.
 
     MuZero is described in the paper:
@@ -85,6 +86,8 @@ class MuzeroRepresentationImpl(OffPolicyAlgorithm):
 
     """
 
+    # codespell:ignore-end
+
     def __init__(
             self,
             observation_spec,

alf/algorithms/taac_algorithm.py

@@ -230,7 +230,7 @@ class TaacAlgorithmBase(OffPolicyAlgorithm):
     In a nutsell, for inference TAAC adds a second stage that chooses between a
     candidate trajectory :math:`\hat{\tau}` output by an SAC actor and the previous
     trajectory :math:`\tau^-`. For policy evaluation, TAAC uses a compare-through Q
-    operator for TD backup by re-using state-action sequences that have shared
+    operator for TD backup by reusing state-action sequences that have shared
     actions between rollout and training. For policy improvement, the
     new actor gradient is approximated by multiplying a scaling factor to the
     :math:`\frac{\partial Q}{\partial a}` term in the original SAC’s actor

alf/summary/render.py

@@ -266,7 +266,7 @@ def is_rendering_enabled():
 def _rendering_wrapper(rendering_func):
     """A wrapper function to gate the rendering function based on if rendering
     is enabled, and if yes generate a scoped rendering identifier before
-    calling the rendering function. It re-uses the scope stack in ``alf.summary.summary_ops.py``.
+    calling the rendering function. It reuses the scope stack in ``alf.summary.summary_ops.py``.
     """
 
     @functools.wraps(rendering_func)

alf/utils/losses.py

@@ -129,7 +129,7 @@ def iqn_huber_loss(value: torch.Tensor,
             is between this and the target.
         target: the time-major tensor for return, this is used as the target
             for computing the loss.
-        next_delta_tau: the sampled increments of the probability for the input 
+        next_delta_tau: the sampled increments of the probability for the input
             of the quantile function of the target critics.
         fixed_tau: the fixed increments of probability, for non iqn style
             quantile regression.
@@ -166,7 +166,7 @@ def iqn_huber_loss(value: torch.Tensor,
     error = loss_fn(diff)
     if iqn_tau:
         if diff.ndim - tau_hat.ndim > 1:
-            # For multidimentional reward:
+            # For multidimensional reward:
             # diff is of shape [T or T-1, B, reward_dim, n_quantiles, n_quantiles]
             # while tau_hat and next_delta_tau have shape [T or T-1, B, n_quantiles]
             tau_hat = tau_hat.unsqueeze(-2)