MLFlow Flashcards
What is MLFlow?
1/A lightweight set of open source tools that improve collaboration and reproducibility of DS work
2/ Use cases : experiment tracking, model registry, MLflow projects
What problems does MLFlow solve?
1/Enables data science collaboration
2/Ensures consistency and reproducibility of ML work
3/Enables automation in ML development
Why do customers care about MLFlow?
1/Ease of Use
2/Open Source
3/Enables Machine Learning operations (ML Ops)
Who should you position MLFlow to?
1/To Data Scientists/ML Engineers who are looking to improve collaboration
2/Customers interested in implementing MLOps
How does MLFlow Work?
1/DataBricks manages the MLflow server
2/MLFlow is seamlessly integrated in Databricks UI
What are the biggest capabilities?
1/ EXPERIMENT TRACKING - Records input,output of each DS experiment allows u to reproduce it
2/MODEL REGISTRY - Central location for a team to store machine learning models, with MLOps and Governance
3/PROJECTS - Captures training dependencies automatically
True or False : MLFlow on Databricks has additional costs?
False. Its free
True or False : Using MLFLow locks you into the platform
False : You can migrate the resources to OpenSource MLFlow
True or False : ML on Databricks is the same as OpenSource MLFlow.
False : Databricks built access controls, webhooks, autoML, Feature Store, Model serving
What should I look for in pitching MLFlow?
1/Growing Data Science teams without a standardized approach
2/Teams struggling to get models in production
3/Organizations interested in improving MLOps
Competitors of MLFlow?
Sagemaker, Comet ML, OpenSource MLFlow, Vertex AI
How much does MLFLow Cost?
Completely Free - customers use MLFlow with clusters to drive DBU consumption.
