Detecting Silent ML Failure with an Open Source Python library - Launch

Over the past year and a half, we have worked closely with design partners across multiple industries with diverse ML use cases.

After lots of blood, sweat, and tears, we developed an algorithm called Confidence-based Performance estimation (CPBE). It allows you to estimate post-deployment model performance, WITHOUT TARGETS/GROUND TRUTH 🤯

It is the only #opensource algorithm capable of fully capturing the impact of data drift on performance.

I'm writing this on launch day as we currently rank #1 product of the day with 115 upvotes and 47 comments. We also just achieved out goal of 300 stars on GitHub a few moments ago and popped our bottle of champaign!

Here's the summary of how we did it:

Choose launch day based on goal: If your goal is to be in the top 5 launch on the weekend, if your goal is reach, launch on a Wednesday

Try to maintain a constant stream of upvotes instead of all upvotes at once. For this share the launch on one social media at a time

Have an active account on PH with 3-5 upvotes

Aim for a 4:1 Upvote/Comment ratio

Prioritise LinkedIn and Twitter

I hope these tips help you!

We’re fully open source and free! Check it out here: https://github.com/NannyML/nannyml.

Here's our Product Hunt if you wanna take a look: https://www.producthunt.com/posts/nannyml.