MLOps focuses on the deployment, testing, monitoring, and automation of ML systems in production. ML engineers use tools for continuous improvement and evaluation of deployed models. They work to enable fast and flexibile deploying the best performing ML models.
This course introduces students to MLOps tools and best practices for deploying, evaluating, monitoring and operating production ML systems on Google Cloud AI Platform.
In addition to UI operations, you also learn extensively how to program and automate MLOps tasks by using Google AI Unified API services
The course offers both theoretical and practical, lab-heavy modules. By completing you would be able to:
Identify and use core technologies required to support effective MLOps.
Implement reliable and repeatable training and inference workflows.
Configure and provision Google Cloud frameworks for reliable and effective MLOps environments.
Integrate ML workflows with overall data workflows to maintain end-to-end ML pipelines
Build key skills for Google professional ML engineers Link to certification
This course is packed with practical exercises and code labs.
* We meet twice a week at online classroom (powered by zoom)
* Practical walkthroughs that present solutions to actual, real-world problems and challenges
* A no-nonsense teaching style that cuts through all the cruft and help you master MLOps on Google Cloud AI Platform
- Session 1: Mar 16, 5pm-7pm PDT (US Pacific Timezone, Daylight Saving Time, GMT-7)
- Session 2: Mar 18, 5pm-7pm PDT
- Session 3: Mar 23, 5pm-7pm PDT
- Session 4: Mar 25, 5pm-7pm PDT
- Session 5: Mar 30, 5pm-7pm PDT
- Session 6: Apr 1, 5pm-7pm PDT
- Session 7: Apr 6, 5pm-7pm PDT
- Session 8: Apr 8, 5pm-7pm PDT
- 4 weeks / 8 sessions / 16 hours
- Lectures / hands-on labs / Projects
- Live session (with zoom) and real time interaction
- Slack support during/after class
Check the Syllabus tab for full course content.
WHO SHOULD LEARN:
Developers, engineers, data scientists
Python coding skills, basic ML knowledges
Est. time spend per week: 4 hours live class (required) + 2 hours labs/projects (optional)
Full refund upon request before the first session ends (Mar 16th, 2021 7pm PDT). 5% transaction fee is not refundable.
If missed live sessions, you can watch recordings any time, along with interactive learning tools, slides, course notes
Students have one year access to course materials
Earn Certificate of Completion
Module 1: ML and MLOps introduction
This session discusses overview on machine learning and MLOps, and how to frame a business problem for ML deployment.
- Machine learning overview
- MLOps overview
- ML end-2-end production pipeline
- Framing a business problem into a e2e pipeline
Module 2: AutoML Image classification
This session will deep dive into MLOps using image classification as an example, teaching how to build end-2-end services, from data ingestion, model training, model deployment, and serving.
- Creating a managed datasets
- Training a managed model
- Deploying a managed model
- Online prediction with deployed model
Module 3: AutoML Image models
This session will dive further into other AutoML image models, object detection and image segmentation, scaling for serving online predictions, and exporting for edge devices.
- Do batch prediction with deployed model
- Create/Train/Deploy object detection model
- Create/Train/Deploy image segmentation model
- Exporting image models to edge devices
Module 4: AutoML Text models
This session will deep dive into AutoML text models: text classification, text sentiment analysis, and entity extraction.
- Create/Train/Deploy text classification model
- Create/Train/Deploy text sentiment model
- Create/Train/Deploy text entity extraction model
Module 5: AutoML Tabular models
This session will deep dive into AutoML tabular models: regression, classification and forecasting. The session will also cover exporting tabular models for custom hosting using TFX serving
- Create/Train/Deploy tabular model
- Feature engineering for tabular model
- Export and Deploy to a custom TFX serving container.
Module 6: Custom training
This session discusses custom training: batch scripts, monitoring, model artifact retrieval, validation, custom training containers, distributed training.
- Creating and executing custom training scripts.
- Monitoring training of a custom training script.
- Retrieving trained model and converting to managed model.
- Validating a trained model.
- Training with custom containers.
- Distributed training.
Module 7: Custom prediction
This session discusses custom prediction: virtualizing deployment, load balancing, custom serving functions, custom prediction containers.
- Pre-built prediction endpoints.
- Load balancing serving requests.
- Designing custom serving functions.
- Prediction with custom containers.
Module 8: Advanced topics
This session discusses more advanced techniques: versioning, A/B testing, local debugging of serving functions, container logging.
- Model versioning.
- A/B testing of production models.
- Container logging.
- Local debugging of custom serving functions.
Andrew Ferlitsch is a machine learning expert at Google. he educates software engineers in machine learning and artificial intelligence. He is the creator of and oversees the development of the open source project Gap, which is a ML data engineering framework for computer vision. Andrew was formerly a principal research scientist at Sharp Corporation, working on imaging, energy, solar, teleconferencing, digital signage, and autonomous vehicles.