Installing SAME

Installing SAME¶

System Requirements¶

Python 3.8+
Docker (optional, used by same verify command)

Install `sameproject` from PyPI ¶

For example, with pip

pip3 install --upgrade sameproject

Verify installation¶

Validate successful installation by running same version. Output should look similar to below

same version

0.2.2

Connecting to a Workflow Engine¶

To run SAME, you will need a workflow engine to connect to. We support Kubeflow Pipelines.

Option A: Test Drive¶

Use a test drive Kubernetes cluster with Kubeflow Pipelines preinstalled. This test drive cluster will expire 1 hour after starting it. You can run testctl get again to get a new one at any time.

Register for a Testfaster account.
Copy and run the testctl install instructions at Access Token. Make sure to include the testctl login command.

Clone this Kubeflow Combinator repo and get a cluster:

git clone https://github.com/combinator-ml/terraform-k8s-stack-kubeflow-mlflow
cd terraform-k8s-stack-kubeflow-mlflow
cd examples/testfaster

testctl get

export KUBECONFIG=$(pwd)/kubeconfig

(If you use a non-bash shell, you may need to spell the command to set an environment variable differently.)

To launch the Kubeflow Pipelines UI, from the terraform-k8s-kubeflow/testfaster directory, run:
```
testctl ip
```
And copy the final URL into your browser. Log in with admin@kubeflow.org and 12341234. Go to Pipelines -> Experiments -> Default in the Kubeflow UI.
Now in the same shell you ran the export command, you can run same run (see next section) and it will be able to find and deploy to Kubeflow pipelines in the configured cluster.

Option B: Use Existing Kubeflow Pipelines Install¶

Ensure your active kubectl context is pointing to the Kubernetes cluster with Kubeflow Pipelines installed:

kubectl config current-context

If so, you are ready to run same run (see next page for example).

After running same run, look in Pipelines -> Experiments -> Default in the Kubeflow UI.

Option C: Install Kubeflow Pipelines on Kubernetes¶

Follow the Kubeflow installation instructions, and then follow the existing Kubeflow Pipelines section.

Or, try the Test Drive if this is too much hard work.

Next¶

You're done setting up SAME and are now ready to execute! The default execution uses Kubeflow (either locally or in the cloud). Use the -t flag to set another target.

To try out an example, check out the roadsigns example from Example Notebook.