How does ML Launchpad differ from Cortex

[schuderer]

Cortex is a new solution to make deploying machine learning solutions easier. What are the similarities and what are the differences to ML Launchpad?

[schuderer]

Great to see that there are more and more solutions coming out in the last few months! People and companies who struggle with operationalizing ML stuff will have a lot of options to choose from! 😃

Some similarities I already see:

• Python-based
• Data Scientist fills in function body (for prediction), getting parameters as arguments, returning the prediction
• Pickled models
• Yaml-based config
• Command line interface
• Open source (Apache 2.0)
• Does not prescribe the libraries to use

Some differences I see at first glance:

• Designed for deployment to AWS (deployment prerequisites are considered out of scope for core ML Launchpad, which means you or your specialists have to solve the deployment pipeline (once), but for this you gain flexibility on where/how to deploy. We are open to suggestions on deployment tools to add to the ML Launchpad ecosystem, however. We try to keep ML Launchpad itself neat and clean, but having tooling as options is certainly possible.)
• No train-test-predict(-retest-retrain)* life cycle support (ML Launchpad also includes model/api versioning, tracking of test metrics)
• Functionality for tracking API calls (ML Launchpad currently only logs them)
• Different APIs in one cortex config (each in a separate config in ML Launchpad)
• No data abstraction builtin (vs. ML Launchpad's data sources, data sinks)
• Batch API (in ML Launchpad you would typically use a data source as batch input. Extending ML Launchpad to support batch predictions with API-inputs would be possible, though. Please create an issue if this would be useful to you)
• No R support
• No file support as far as I can see
• A company's product (Not sure how that affects whether what is open-sourced is actually the complete product or there some tools/parts that are still closed-source. I think the product is 100% open source and the monetization part is via support and consulting.)

I think it would actually be possible to combine Cortex with ML Launchpad, combining their advantages, and I plan to try it out when I find the time.