Published On May 28, 2021
This is a FULL TUTORIAL that has 2 Parts. First, we interview Special Guest: Will Landau, Creator of Targets! Then we do an insane forecasting lab implementing #modeltime & #targets to:
- Make 32 Time Series Models (18 ARIMA, 18 Prophet)
- Track Accuracy for every model/time series combination
- Select best models for each of 18 time series
- Automates a Forecast Audit with Error Reporting
WANT THE CODE?
Join Learning Labs PRO: https://university.business-science.i...
WANT TO LEARN TIME SERIES FORECASTING?
Join my Time Series Course: https://university.business-science.i...
TABLE OF CONTENTS:
00:00 Energy Forecasting: Modeltime & Targets
00:50 Goals for Today: Targets for Production Forecasting
02:35 Workflow: Modeltime - Targets - Rmarkdown
04:52 Interview with Will Landau, Creator of Targets
05:35 In Grad School, PhD Work: Models with Long Runtimes
07:02 Identified a Gap in R Ecosystem: No Pipeline Tools
07:37 End of Grad School began developing Drake (& then Targets)
09:25 Why working on Targets gives Will joy
09:40 R Community: How it's benefited Will's R Package Development
10:10 ROpenSci: Access to the best developers in R community
10:46 Kirill Mueller's Influence: Proposing High-Performance Computing
11:40 What is ROpenSci?
12:59 Matt & Will's Shared Experience with R Community
14:29 Will's Bayesian & Statistics Background
15:00 Iowa State: BioTech NextGen DNA Sequencing Data Analysis Group
16:12 Genomics Project: GPU Computing, Hierarchical Models, & Genomics Data
16:47 Modeling Crop Yield with Genomics (Massive Models)
17:30 STAN & JAGS Models were too computationally expensive
18:06 Creating a Markov Chain Monte Carlo (MCMC) Simulation using GPUs
18:44 Massive Speed Gains: Turned Days (CPU) to 4 Hours (GPU)
19:15 Will wishes he had Targets: Instant Parallelization
20:20 Parallel Computing is Simple in Targets
21:30 FREE RESOURCE: Targets Book https://books.ropensci.org/targets/
23:07 Business Problem: Scalable Time Series Modeling with ARIMA & Prophet
26:21 Forecast Audit Report (Data Product)
28:12 Why Targets?
32:00 Key Concept: Branching https://books.ropensci.org/targets/dy...
34:00 Tarchetypes: Targets Ecosystem Expansion https://docs.ropensci.org/tarchetypes/
36:30 Code Demo: Targets + Modeltime
36:48 Targets Workflow for Energy Forecast Reporting
38:04 Project Setup
39:45 Module 01: Targets Branching Basics
43:01 Branching with tarchetypes::tar_group_by()
48:38 Dynamic Modeling: 15 Linear Regresions by Auto Manufacturer
53:17 Broom Tidiers: Getting Coefficient & Accuracy Metrics for 15 LM Models
57:50 Module 02: Time Series Forecasting with Modeltime + Targets
1:01:30 Data Import & Preparation Targets
1:06:03 Clean Energy Data Target
1:07:37 Extend Energy Data Target
1:10:40 Branching to 18 Time Series with tarchetypes::tar_group_by()
1:13:46 Time Series Splitting
1:15:41 Making 36 Time Series Models: 18 ARIMA & 18 Prophet
1:18:37 LL PRO Challenge: Add a GLMNet Model
1:18:56 Test Set Accuracy & Model Comparison
1:23:13 Model Selection (Lowest RMSE)
1:25:08 Model Refitting
1:26:58 Final Forecast (Future Data)
1:29:37 Forecast Audit (Accuracy Checking)
1:32:56 Automated Report
1:34:53 Learning More: 5-Course R-Track Program https://university.business-science.i...
1:41:14 Q&A