This install script is an example of how to get the necessary dependencies, as well as pydreamer and NavDreams code and python libraries installed in development mode. It was used on ubuntu 18.04 machines for training.
Download the navdreams_data
folder to your home directory so that ~/navdreams_data/wm_experiments is a valid location. (Note: we have reports that GDrive may be adding unwanted extensions to files, such as adding '.zip' to model checkpoints. We will investigate and switch to another hosting solution, until then, if you get an error that a file is missing, check the extension)
Note that this is quite a large download, you can potentially only download the parts which you requre to save on space.
(Alternatively, you can generate the datasets using the simulator.
Afterwards, make sure to move them to the location expected by the script e.g. ~/navdreams_data/wm_test_data)
python make_vae_dataset.py --scope SCR
In test_n_step dream, dream sequences are generated by world-models, and then compared using pixel-wise error, as well as plotted for visualization.
First, make sure you source the virtualenv
source ~/n3denv/bin/activate
cd ~/Code/NavDreams
Now you should be able to run the evaluation script
python test_n_step_dream.py
Which outputs the dream predictions, and errors for a couple of example sequences
Running with the flag --error computes the same error on the whole dataset
python test_n_step_dream.py --error
Other flags (for example --gifs, allowing to create dream gifs) can be found directly in test_n_step_dream.py