3. Set up a new RAMP event¶

3.1. Deploy a specific RAMP event¶

Now we want to deploy a specific problem and event. We will demonstrate by by deploying an iris event.

First, you need to get the starting kit and the data into the ramp_deployment directory:

~/ramp_deployment $ mkdir ramp-kits
~/ramp_deployment $ mkdir ramp-data
~/ramp_deployment $ git clone https://github.com/ramp-kits/iris ramp-kits/iris
~/ramp_deployment $ git clone https://github.com/ramp-data/iris ramp-data/iris
~/ramp_deployment $ cd ramp-data/iris
~/ramp_deployment/ramp-data/iris $ python prepare_data.py
~/ramp_deployment/ramp-data/iris $ cd ../..

Next, you need to create a configuration file for a specific event. You can create this configuration file by executing the following command line from the ramp_deployment folder:

~/ramp_deployment $ ramp setup init-event --name iris_test

The above creates a events/iris_test directory inside the deployment directory, and populates it with a config.yml with the configuration specific to the event.

The config.yml file should look like:

ramp:
    problem_name: iris
    event_name: iris_test
    event_title: "Iris event"
    event_is_public: true
worker:
    worker_type: conda
    conda_env: ramp-iris
dispatcher:
    hunger_policy: sleep
    n_workers: 2
    n_threads: 2
    time_between_collection: 1

ramp configuration

problem_name: All events using the same problem should have the same problem_name.
event_name: Must always be preceded with ‘<problem_name>_’ as in the example above.
event_title: Human readable event name to display on website.
event_is_public: Boolean indication whether or not the event should be public.

worker configuration

worker_type: Refer to the documentation about workers for more information on worker types available.
conda_env: Name of the conda environment to use. If not specified, the base environment will be used. Only relevant if using a conda worker.

dispatcher configuration

hunger_policy: Policy to apply in case that there is no anymore workers to be processed. One of None, ‘sleep’ or ‘exit’, default is None:
- if None: the dispatcher will work without interruption;
- if ‘sleep’: the dispatcher will sleep for 5 seconds before to check for new submission;
- if ‘exit’: the dispatcher will stop after collecting the results of the last submissions.
n_workers: Maximum number of workers that can run submissions simultaneously.
n_threads: The number of threads that each worker can use.
time_between_collection: How long, in seconds, the worker should wait before re-checking if the submission is ready for collection. The default is 1 second.

Before you continue make sure that:

Your private and public data are stored in the correct folders.

For example, your data might be in:

public data: ramp_deployment/ramp-kits/<problem_name>/data/, and

private data: ramp_deployment/ramp-data/<problem_name>/data/.

Note:

for more information on directory structure of the starting kits see Overall directory structure
The conda environment (set in the config.yml file above, here called ramp-iris) used by your event exists. Note, that this only applies to events that use the conda worker. To check which conda environments are available, use:
$ conda env list

Now, you can easily deploy the event (adding both problem and event to the database) by calling from the deployment directory:

~/ramp_deployment $ ramp setup deploy-event --event-config events/iris_test/config.yml --no-cloning

Without passing --no-cloning, it will try to clone the starting kit and data from the https://github.com/ramp-kits/ GitHub organization. If the starting kit is located on this GitHub organization, you can skip the first step above, which manually clones the kit and prepares the data, and let ramp setup deploy-event do this for you.

3.2. Launch the dispatcher to train and evaluate submissions¶

At this stage, you can launch the RAMP dispatcher from the ramp_deployment directory, which will be in charge of training, evaluating submissions, and updating the database:

~/ramp_deployment $ ramp launch dispatcher --event-config events/iris_test/config.yml -vv

If you are running the dispatcher on a remote server, you want to launch it within a terminal multiplexer as screen or tmux. It will allow you to detach the process and let it run. Refer to the documentation of screen or tmux to use them.

3.3. Launch several dispatchers at once¶

In case that you are running multiple events in parallel, you will want to start several dispatchers, on for each open event. We provide a daemon which will be in charge of managing the pool of dispatchers. You can start it as:

~/ramp_deployment $ ramp launch daemon --events-dir events --verbose

To can interrupt the daemon by pressing the combination of keyboard keys Ctrl+C. You can start launch the daemon within tmux or screen as well.