sqlflow
sqlflow copied to clipboard
sql-machine-learning
Make SQLFlow easy to install / use
Is your feature request related to a problem? Please describe. As for now, we have a split our all-in-one docker image into multiple small images. Each module's responsibility is more clear. However, the total size of these images get larger. Also, to start SQLFlow, we have to do more things than before. Our playground solve this problem in some way, but it relies on VM. What is the future release form? How can we make it easy to use like the old all-in-one package? We need to discuss this.
Describe the solution you'd like Easy to download / install and run SQLFlow.
Describe alternatives (optional)
Use our playgorund
Additional Notes
I'd like to have the all-in-one Docker image back. For user quick start purpose only. And the current "local mode" is easy to maintain. For using custom model images like TO TRAIN hub.docker.com/sqlflow/some_model_image:v0.1 we can document to let users try this out on a cloud playground or install Kubernetes by themselves.
We'd better make a series of tools. Like playground, a single yaml file, all-in-one docker image, etc. The all-in-one docker image should only for quick start. However, it does not support cluster mode. I'm not sure if it is reasonable to provide it.