Launching an Experiment from the Virtual Coach

Starting the Virtual Coach

Local source installation

In case you have installed NRP from the source, there is a special alias for running the Virtual Coach. For this alias to work, $HBP/user-scripts/nrp_aliases has to be sourced in your bash.rc. The configure_nrp script takes care of that automatically. The Virtual Coach alias is cle-virtual-coach and can be run in three different ways:

There are different ways to start a Virtual Coach instance. With a local NRP install you can use the alias cle-virtual-coach in three different ways:

  1. Launch a Jupyter Notebook session

    To launch a Jupyter Notebook session just run cle-virtual-coach jupyter notebook in a terminal. Of course, Jupyter Notebook has to be installed prior.

  2. Launch an interactive python interpreter session

    To launch an interactive python interpreter session just run cle-virtual-coach python in a terminal.

  3. Launch a python script with arguments

    To launch a python script foo.py with arguments a b c just run cle-virtual-coach foo.py a b c.

This information is also available in the alias help cle-virtual-coach -h.

Local Docker installation

You can also use Virtual Coach with the local Docker installation. In order to do that, you should follow these steps:

  1. Install the latest NRP Docker version with the command of ./nrp_installer.sh install latest the script exposes port 8888 by default for the Jupyter.

  2. Run ./nrp_installer.sh connect_backend which opens a terminal inside backend container.

  3. Run the command of cle-virtual-coach jupyter notebook "--ip=0.0.0.0" inside backend container to start Jupyter.

This will give you a URL like http://localhost:8888/?token=f027e20b23dd04d34ec827c6ff8ee402ccdf4c775d440766 that you should open with your browser.

If you want to specify another port for Jupyter, the steps above change as follows:

  1. When installing nrp you need to add -np/--notebook_port <port_number> to install command. For example, the command of ./nrp_installer.sh -np 9875 install latest will install nrp with port 9875 to be exposed for Jupyter.

  2. Run ./nrp_installer.sh connect_backend which opens a terminal inside backend container.

  3. Run the command of cle-virtual-coach jupyter notebook "--ip=0.0.0.0" --port=9875 inside backend container to start Jupyter on port specified in step 1. This command yields a URL that allow you to access the Jupyter notebook run on the container.

To test that the installation was successful, create a new notebook and run this line of code:

from pynrp import virtual_coach

if you get no error running this cell, your Jupyter notebook is ready.

Launching a Simulation

The first thing we need to do is to import the Virtual Coach and create a VirtualCoach instance connected to an NRP server.

from pynrp.virtual_coach import VirtualCoach
vc = VirtualCoach(environment='http://localhost:9000', storage_username='nrpuser', storage_password='password')

Here we chose the local machine as the NRP server. The Virtual Coach can connect to NRP servers running in other locations by setting the environment parameter to the proper URL. For example, it is possible to connect to the NRP server running at https://nrp-legacy.neurorobotics.ebrains.eu by executing this command:

vc = VirtualCoach('https://nrp-legacy.neurorobotics.ebrains.eu', oidc_username='<your-hbp-username>', oidc_password='<your-hbp-password>')

Notice that parameters storage_username and storage_password are replaced with oidc_username and oidc_password. The latter must be used when EBRAINS OIDC authentication is required, the former are used otherwise.

Note

Prior using Virtual Coach with the online NRP servers, you should login at least once to that online server with your EBRAINS account in order to give the user consent.

Once we created a VirtualCoach instance, we can use it to check out the current state of our environment. We can see a list of the available experiments to run, a list of the available servers to run experiments on, and a list of the currently running experiments.

vc.print_templates()
vc.print_available_servers()
vc.print_running_experiments()

After making sure our experiments exist and enough resources are available, we can attempt to launch an experiment. In this next section we will launch the Template Husky experiment. We first need to clone the experiment in our storage. To launch a specific experiment, we need to specify its configuration name, which we can get from the cloned experiment list.

exp_id = vc.clone_experiment_to_storage('ExDTemplateHusky')
vc.print_cloned_experiments()
sim = vc.launch_experiment(exp_id)

Launching an experiment can take some time and once it’s been successfully launched, we’ll get a Simulation Successfully Created log.

We can also make sure that the experiment has been launched by querying the Virtual Coach for currently running experiments.

Read more about the interaction with the experiment throught the Virtual Coach in the following tutorial.