Productionizing your forecast model

Forecasting in Practice

Rami Krispin

Senior Manager, Data Science and Engineering

Introduction

  • Senior manager - data science and engineering
  • Time series, forecasting & MLOps
  • Author
  • Open source contributor

Forecasting in Practice

Forecasting in Production

  • Forecasting pipeline architecture
  • Set automation
  • Monitor the pipeline

The US Hourly Demand Forecast

Forecasting in Practice

Motivation

  • Automation - a recurring forecast task
    • Example: forecasting the hourly temperature
  • Scale - large amount of series
    • Example: many series or high computing needs
Forecasting in Practice

General architecture

Forecasting Pipeline

Forecasting in Practice

General architecture

Forecasting Pipeline

Forecasting in Practice

General architecture

Forecasting Pipeline

Forecasting in Practice

General architecture

Forecasting Pipeline

Forecasting in Practice

General architecture

Forecast Pipeline

Forecasting in Practice

General architecture

Forecasting Pipeline

Forecasting in Practice

Course outline

Chapter 1

  • Review data source
  • Pull the data from the EIA API.

Chapter 2

  • Experimentation process
    • Train, test, and log multiple forecasting models

Chapter 3

  • Deployment process
    • Data automation, model refresh & capturing logs

Chapter 4

  • Post-deployment steps
    • Monitoring & alerts
    • Best practices
Forecasting in Practice

Course prerequisites

Advanced course, prior knowledge required:

  • Time series analysis and forecasting
  • Orchestration systems
  • APIs
  • Python
Forecasting in Practice

Course tools

ML system architecture including an ETL pipeline, experiment framework, ML pipeline, and dashboard

Forecasting in Practice

Let's practice!

Forecasting in Practice

Preparing Video For Download...