Seamless Deployment of the models to Production
Rohit Adhikari (~rohit40) |
Seamless Deployment of the ML models to Production:
As the major challenge in Machine Learning is to deploy your model to production. Most of the models were developed after the research done by Data Scientist and ML Engineers in their local system and when they plan to deploy it in production they face a lot of challenges in integrating the models with the application, retraining of models, monitoring of models, versioning of models, etc. This is the reason 80-85% of the model doesn't go into production.
MLOps is the solution to overcome this issue as it empowers the data scientist and software engineer to bring the ML models to Production. It also enables you to track, version, audit, re-use every stage in your Machine Learning lifecycle, and provide orchestration services to streamline this workflow.
How MLOps is different from DevOps?
Data/model versioning != code versioning, how to version the data sets as the schema and origin data change
Model reuse is different from software reuse, as models must be re-tuned based on input data/scenario.
To reuse a model you may need to fine-tune on it (meaning you need the training pipeline)
Models tend to decay over time & you need the ability to retrain them on demand to ensure they remain useful in a production context.
DevOps = Development + Operations
MLOps = Machine Learning(Data and Model) + Development + Operations
Key Challenges to solve with MLOps?
Model reproducibility & versioning:
Track, snapshot & manage assets used to create the model
Enable collaboration and sharing of ML pipelines
Model audibility & explainability:
Maintain asset integrity & persist access control logs
Certify model response meets regulatory & adversarial standards
Model packaging & validation:
Support model portability across a variety of platforms
Certify model performance meets functional and latency requirements
Model deployment & monitoring:
Release models with confidence
Monitor & know when to retrain by analyzing signals such as data drift
How to build your MLOps platform?
Collaboration Tool + Development & Deployment Tools + Monitoring and Operations Tools
Knowledge of Machine Learning life cycle Understanding of dockers to containerize the image Basic knowledge of Kubernetes Understanding of Kubeflow - Machine Learning workflow to develop, deploy and serve the machine learning models
I'm a Sr. Data Scientist, worked on different projects Machine learning, Deep Learning, Computer Vision, and MLOps. Passionate about building intelligent machines using AI ML and deploy them seamlessly to the productions so that it can be easily used for inference. Holding a Masters Degree in Computer Application and PGP in Big Data and Machine Learning.