###################################
TRAINING
##################################

To train a model you need to run the main.py file using the following parameters
--dataset_type 'dataset' --latent_algo 'algo' 

where 
	-'dataset' can be expert, medium or expert_replay
	-'algo' can be lompo, combo, oraac, lodac

By default, we use a trained latent model and an already computed latent_real_dataset
(which is the inference of the dataset into the latent sapce) and the environment is stochastic.
We can change these parameters using --load_pretrained_latent, --load_latent_real_dataset and --risky_env.



##################################
EVALUATIONS
##################################

Our models are saved in the logdir folder (with the path corresponding to the training parameters). To evaluate these models, we can use the same
parameters as above, setting --evaluate to True.
By default, we evaluate our model for 100 episodes. This can be changed using --num_evaluations.
For this evaluations, the models in the folder called 'algo' are loaded. 


##################################
CREDITS
##################################

Some part of this code has been borrowed from the official github of LOMPO (Latent Offline Model-based Policy Optimization).



