AI Workflow: Enterprise Model Deployment
- Deploying Models
- Today data scientists have more tooling than ever before to create model-driven or algorithmic solutions, and it is important to know when to take the time to make code optimizations. This week we spend a lot of time performing hands on activities. We start this week by interacting with Apache Spark then progressing to a tutorial with Docker. We’ll wrap up the week working through a tutorial on Watson Machine Learning.
- Deploying Models using Spark
- This week is primarily focused on deploying models using Spark. The rationale to move to Spark almost always has to do with scale, either at the level of model training or at the level of prediction. Although the resources available to build Spark applications are fewer than those for scikit-learn, Spark gives us the ability to build in an entirely scaleable environment. We will also look at recommendation systems. Most recommender systems today are able to leverage both explicit (e.g. numerical ratings) and implicit (e.g. likes, purchases, skipped, bookmarked) patterns in a ratings matrix. The majority of modern recommender systems embrace either a collaborative filtering or a content-based approach. A number of other approaches and hybrids exist making some implemented systems difficult to categorize. We wrap the week up with our hands-on case study on Model Deployment.