rlberry.manager.tensorboard_to_dataframe

rlberry.manager.tensorboard_to_dataframe(tensorboard_data)

Function to convert ‘tensorboard log’ to ‘Panda DataFrames’.

To convert the ‘tensorboard log’, the input must be must be the path to “the parent folder of all the training log” (path_to_tensorboard_data), and the ‘events.out.tfevents’ files have to be in this kind of path :

< path_to_tensorboard_data/algo_name/n_simu/events.out.tfevents.xxxxx >

Or you can specify all the desired ‘events.out.tfevents’ with a Dict. In that case, the key should be the algorithm name, and the value the list of the ‘events.out.tfevents’ path. The seed/n_simu number wille be the position in the list.

The output format is a dictionary.

key = tag (type of data)

value = Panda DataFrame with the following structure (4 column):

• “name” = algo_name

• “n_simu” = n_simu (seed)

• “x” = step number

• “y” = value of the data

Parameters:

path_to_tensorboard_data (str or Dict):

if str: path to the parent folder of the tensorboard’s data. if dict: Key = algo_name , value = list of ‘events.out.tfevents.xxxxx’ path

Returns:

Dictdict of Panda DataFrame (key = tag, value = Panda.DataFrame)