There are various challenges in MLOps and model sharing, including, security and reproducibility. To tackle these for scikit-learn models, we've developed a new open-source library: skops. In this article, I will walk you through how it works and how to use it with an end-to-end example.
One piece of structural discrimination: When men who have failed at their agreed-upon goal are given more resources & people to keep working on it; while women who have succeeded at their agreed-upon goal are told they haven't succeeded and the goal is actually different now.
The first part is a type of "failing upwards". The second "shifting goalpost".
Having language to talk about the patterns helps to make them a bit more apparent, I believe, and ultimately can change the system.
When versioning your experiments, it's best to keep couple of information for better reproducibility:
🧑🏻🔬 Your hyperparameters and attributes of preprocessors and architecture (pipeline)
📈 Metrics
👑 Feature importances (which I used ELI5 for)
✅ Requirements of your environment
Worble Parker
Margaret Mitchell