README.md

In hoxo-m/prophetExt: Extension for Facebook Prophet

prophetExt

Extension for Facebook Prophet.

Installation

You can install prophetExt from github with:

# install.packages("devtools")
devtools::install_github("hoxo-m/prophetExt")

or

# install.packages("githubinstall")
githubinstall::githubinstall("prophetExt")

Example

Ready data.

df <- read.csv("https://raw.githubusercontent.com/facebook/prophet/master/examples/example_wp_peyton_manning.csv")
df$y <- log(df$y)

Fit model.

library(prophet)
m <- prophet(df, daily.seasonality = FALSE)

Change Points

Pick changepoints.

library(prophetExt)
cpts <- prophet_pick_changepoints(m)
head(cpts)
#>   changepoint  growth_rate      delta
#> 1  2007-12-10 -0.360276798         NA
#> 2  2008-10-11 -0.002125965  0.3581510
#> 3  2009-01-14  0.458734213  0.4608602
#> 4  2009-10-26  0.217411387 -0.2365794
#> 5  2010-01-31 -0.034489145 -0.2519005
#> 6  2011-02-16  0.252856713  0.2873459

Draw changepoints.

future <- make_future_dataframe(m, 365)
fore <- predict(m, future)
plot(m, fore) + autolayer(cpts)

Outliers

Detect outliers.

outliers <- prophet_detect_outliers(m)

head(outliers)
#>           ds        y     yhat    resid      p_value
#> 1 2008-01-21 10.89720 9.257430 1.639772 4.921098e-02
#> 2 2008-02-04 12.09746 9.218465 2.878991 7.197366e-07
#> 3 2008-02-05 10.63528 8.932996 1.702283 3.401723e-02
#> 4 2009-01-04 10.15382 8.304355 1.849463 1.626819e-02
#> 5 2009-11-16 10.55870 8.868621 1.690078 3.651931e-02
#> 6 2010-01-17 11.00793 9.264015 1.743917 2.793262e-02

Draw outliers.

plot(m, fore) + autolayer(outliers)

Draw outliers using calendar plot.

prophet_calendar_plot(outliers)

SHF (Simulated Historical Forcast)

shf <- prophet_shf(m, periods = 365)

head(shf$estimated)
#>   x      value
#> 1 1 0.05239550
#> 2 2 0.05211918
#> 3 3 0.05184478
#> 4 4 0.05157229
#> 5 5 0.05130174
#> 6 6 0.05103313

plot(shf)

hoxo-m/prophetExt documentation built on Jan. 18, 2021, 11:12 a.m.