watson-train-serve

Training and Serving Models with Watson Machine Learning

This pipeline runs training, storing and deploying a Tensorflow model with MNIST handwriting recognition using IBM Watson Studio service. This example is originated from Kubeflow pipeline's ibm-samples/watson example.

Prerequisites

• Install KFP Tekton prerequisites

• Install Watson Requirements

Instructions

1. Compile the Watson ML pipeline. The kfp-tekton SDK will produce a Tekton pipeline yaml definition in the same directory called watson_train_serve_pipeline.yaml.

python watson_train_serve_pipeline.py

2. If your default Kubeflow service account dosn't have edit permission, follow this sa-and-rbac to setup.

3. Next, upload the watson_train_serve_pipeline.yaml file to the Kubeflow pipeline dashboard with Tekton Backend to run this pipeline. Then, use the default pipeline variables except for these two variables.

   GITHUB_TOKEN: your github token

   CONFIG_FILE_URL: your configuration file which stores the credential information, here is the example of creds.ini file