In this blog post, we will demonstrate how to create a new experiment in the Skymind Intelligence Layer (SKIL) by importing a Zeppelin notebook.
Experiments in SKIL are different configurations of deep learning models and data pipelines, applied to a particular problem. They can be contained in Zeppelin notebooks, which allows us to modify, save, and run code as needed.
SKIL enables developers to quickly prototype and deploy the best models to production to build deep-learning apps. That process starts in a notebook and ends with the SKIL model server.
For instance, SKIL can be used to easily track multiple variants or versions of a model, compare performance between different models, and deploy the best model to production using SKIL’s model server.
Setting up a workspace in SKIL
Before importing a notebook, we must first have a workspace set up in the SKIL user-interface. A workspace is a place to gather and conduct different experiments related to a specific project. To create a workspace, first click on the workspaces tab on the left side of the interface as shown in the image below.
Next, simply click on the “Create Workspace” button on the right. This will bring up a new window as shown below.
You can then name the workspace to distinguish it from other projects and choose the model history server ID. The labels are optional and can be used to help identify a workspace. For this post, we will use the name “First workspace” and give it the label of “first.”
Importing a notebook in a SKIL workspace
Now that a workspace is available, we can import a Zeppelin notebook as an experiment in the workspace. If you click on the workspace tab, you should see the workspace “First workspace” that we created earlier.
Click on the new workspace to see the details of the workspace below.
To import the notebook, click on the “Create New Experiment” button, which will bring up a new window. Select the option “Choose an existing notebook” and click on the button that says “Choose JSON file.”
Finally, select the Zeppelin notebook on your local device that you want to import and fill in the experiment name, model history server ID and Zeppelin server ID. For this post, we will select the experiment name “Parallel Training” and import a Zeppelin notebook “Parallel Training.json.”
Once you click “Create Experiment Notebook” at the bottom, the Zeppelin notebook should be created and a new experiment should be listed in the workspace.
There. We successfully imported a Zeppelin notebook as an experiment using SKIL, which concludes this blog post.
Subscribe to Deep Learning for Enterprise
Get the latest posts delivered right to your inbox