Chris Campbell
Gaussian Processes
Forecasting Processes
Agenda
Gaussian Processes
Profit
Code Speed
Forecasting Processes
Project Structure & Docs
Packaging & Testing
Reproducible Data
Raw Data
1 run_all.R
CommitScripts
f34d20 sales
Data Checksum
02ec42
2 run_all.R 932a45 sales 563b3e
3 run_all.R 932a45 sales 74a10d
Agile Planning & Tickets
Gaussian Processes
Joint Distribution
Conditional Probability
Distribution Next Point
package:GPfit
library(GPfit)gp2 <- GP_fit(
X = xm1, Y = y1)
Kernel
𝐾(𝑥, 𝑥) =
𝑘(𝑥1, 𝑥1 ) 𝑘(𝑥1, 𝑥2 ) 𝑘(𝑥1, 𝑥𝑛 )𝑘(𝑥2, 𝑥1 ) 𝑘(𝑥2, 𝑥2 ) 𝑘(𝑥2, 𝑥𝑛 )𝑘(𝑥𝑛, 𝑥1 ) 𝑘(𝑥𝑛, 𝑥2 ) 𝑘(𝑥𝑛, 𝑥𝑛 )
Slow in High Dimensions
What is taking longest?
library(profvis)profvis(gp2 <- GP_fit(X = xm1, Y = y1))
Timing Hierarchy by Function
Replace Slower Code
Forecasts aggregatedmonthly
https://builder.r-hub.io/
Gaussian Processes
Derive non-linear relationships from observed data
Excellent for interpolation
Allow estimation of forecast error
Poor performance for higher dimensionality