[Notes] Introduction to Machine Learning in Production, Week 2

My notes from MLOps Specialization on Coursera

Course 1, Introduction to Machine Learning in Production, instructed by Andrew Ng (DeepLearning.AI)

Week 1 Notes, Overview of the ML Lifecycle and Deployment

Week 2 Notes, Select and Train a Model

Week 3 Notes, Data Definition and Baseline

Table of Contents

1. Select and train model
2. Error analysis and performance auditing
3. Data iteration

1. Select and train model

1.1 Key challenges

AI system = Code + Data (algorithm/model)

Model development is an iterative process

When building an ML model, there are three key milestones that projects should aim to achieve:

1. Training Set Performance: The first milestone is to ensure that the model performs well on the training set. If the model cannot perform well on the training set, it is unlikely to perform well on other datasets.
2. Dev/Test Set Performance
3. Business Metrics and Project Goals: While achieving high accuracy or performance on the development set is important, it is equally…