Sergio Guadarrama

Can you make sure that your TF version has tf.gfile.Open, tf.gfile.Exists?

You can use `gin.add_config_file_search_path('path/to/config.gin')` to add **folders** that you want gin to use when doing `gin.parse_config_files_and_bindings(...)`

Currently, you can assign values to parameters in code using `gin.bind_parameter` ``` value = ... gin.bind_parameter('my_function.parameter', value) ``` We are thinking in being able to create configurations as Python objects,...

In general defaults should be values that work well by "default" so in most cases work well if they are not modified. In other cases there are specific values that...

@skye having a jax function to do this would be great, even if it has to happen right at the beginning. I can even warn the user if request more...

The specs don't match PolicyStep(action=., state=(), info={'dist_params': {}, 'value_prediction': .}) vs. PolicyStep(action=., state=(), info=DictWrapper({'dist_params': DictWrapper({'logits': .}), 'value_prediction': .})) Make sure the policy builds the correct info data.

I will recommend taking a look at the [Bandit Colab](https://github.com/tensorflow/agents/blob/master/docs/tutorials/bandits_tutorial.ipynb) Double check the `action_spec` and the `action` To add multiple trajectories to the replay buffer increase its size `max_length=N`

You can use `policy_loader.load()` to load a saved policy and evaluate it. See `policy_loader_test.py`