Deploying a model with SKIL
Before we dive into the SKIL UI, we'll run a little Python script to train a Keras model, save it to file and deploy it to SKIL:
This scenario is about SKIL's UI, but if you're interested in what exactly the script does, click on it in the editor to the upper right. Here's how SKIL works on a high level:
Each SKIL task starts out by defining a workspace, where you and your colleagues do all your work. Think of a workspace as a project.
Within a workspace you define one or several experiments in which you run machine learning tasks. In an experiment you can define several models and once you're happy with the performance of a model, you can create a deployment. You then deploy your model to obtain a service. The
train_and_deploy.py script does all of that for you, here is the key part:
skil_server = Skil() work_space = WorkSpace(skil_server, name="My workspace") experiment = Experiment(work_space, name="Keras experiment") model = Model('model.h5', model_id="keras", experiment=experiment) deployment = Deployment(skil_server, name="My deployment") service = model.deploy(deployment)
You now have a deployed SKIL service at your disposal, let's see what that looks like in the SKIL UI!