Merge of PR tensorflow#1443

PiperOrigin-RevId: 233087778

Author: blazejosinski

tensor2tensor/rl/ppo_learner.py

@@ -351,7 +351,7 @@ def initialization_lambda(sess):
         batch_env.initialize(sess)
 
     memory = [
-        tf.get_variable(
+        tf.get_variable(  # pylint: disable=g-complex-comprehension
             "collect_memory_%d_%s" % (epoch_length, name),
             shape=[epoch_length] + shape,
             dtype=dtype,

tensor2tensor/rl/trainer_model_free.py

@@ -62,8 +62,10 @@ def initialize_env_specs(hparams):
 
   return rl.make_real_env_fn(env)
 
+
 step = 0
 
+
 def train(hparams, output_dir, report_fn=None):
   """Train."""
   env_fn = initialize_env_specs(hparams)
@@ -127,6 +129,7 @@ def evaluate_on_new_model(model_dir_path):
                 epoch=0,
                 model_save_fn=evaluate_on_new_model)
 
+
 def main(_):
   hparams = trainer_lib.create_hparams(FLAGS.hparams_set, FLAGS.hparams)