kubeflow_example

This repo is made for kubeflow example

How to use it?

• I wrote an explain on my blog
  • iris : https://lsjsj92.tistory.com/581
  • titanic : https://lsjsj92.tistory.com/586
  • metrics_evaluation_and_check_condtion : https://lsjsj92.tistory.com/589

1. iris

• Make kubeflow pipeline for simple example : iris data
• Doing load data -> training data
• If you want to start this example do this:
  1. Make each dokcer image(preprocessing, training)
  2. Make kubeflow pipeline(pipeline.py)
  3. Start kubeflow
  4. Upload pipeline and start experiment

2. titanic

• Make kubeflow pipline for simple AWS example with titanic data

• The process of this kubeflow pipeline : In preprocessing

  1. Get data from AWS s3
  2. Preprocessing data
  3. Upload to AWS s3 after preprocessing data Until here, doing in preprocessing

  In train_model 4. Get preprocessing data from s3 5. Training model 6. Upload model to s3

3. metrics_evaluation_and_check_condtion

• Make kubeflow evaluation and check condition
• Use iris data
• Evaluation
  • Make metrics json
  • Print metrics in kubeflow
  • in 2_model_training/training_model.py
• Check condition
  • Use dsl.condition
  • in pipeline.py
• Result
  •