Facebook Prophet Slide Deck_FB Prophet Slide Deck

ChristineCheong5 88 views 8 slides Aug 04, 2024
Slide 1
Slide 1 of 8
Slide 1
1
Slide 2
2
Slide 3
3
Slide 4
4
Slide 5
5
Slide 6
6
Slide 7
7
Slide 8
8

About This Presentation

Facebook+Prophet


Slide Content

Facebook Prophet

StructuralTime Series
DescriptionVisualization
StructuralTime Series isthe
decompositionofthedatain at least:
Data Seasonality
Trend Exogenousimpacts
Methodologicalframework
Trend
Seasonality
Exogenousimpacts
Error Term

Facebook Prophet quick facts
Which?
Description
Prophet is customizable in ways that are intuitive to
non-experts
4
Stan background-probabilisticprogramming
languageforstatisticalinference
2
Builtbyfacebook1
Dynamic Holidays3
Built-in Cross Validation5

Prophet
Mechanics
Trend +
Seasonality+
Holiday effects+
External regressors+
error
c(t)
s(t)
h(t)
x(t)
e
Methodologicalframework
Where:

Dynamic Holidays –Valentine‘sexample
Chocolate
demand
11121314
February
YoustateValentine‘sasa key
eventand specifyhowmany
daysbefore/after
Youmustcreatedummy
variables foreachday, ifyou
believetheyhavedifferent
impacts
15
Visualization
Facebook Prophet
Other models:

Facebook Prophet Model
DescriptionComponent
HolidaysHolidays Dataframe thatwepreparedDataframe thatweprepared
SeasonalitySeasonality Yearly, weeklyordaily. True orFalseYearly, weeklyordaily. True orFalse
Seasonality_modeSeasonality_mode MultiplicativeoradditiveMultiplicativeoradditive
Seasonality_prior_scaleSeasonality_prior_scale Strength of the seasonalityStrength of the seasonality
Holiday_prior_scaleHoliday_prior_scale Larger values allow the model to fit larger seasonal fluctuationsLarger values allow the model to fit larger seasonal fluctuations
Changepoint_prior_scaleChangepoint_prior_scale Does the Trend change easily?Does the Trend change easily?

Pros and Cons
Can needintenseoptimization
2
Not goodwithshort-term dynamics
3
Complexprogramming
1
Flexible
1
Built-in Cross Validation
2
Dynamics Events
3
Great withRegressors
4

Challenge
A Dataset withDaily UdemyWikipedia Visits
Description
PredictthenumberofvisitstotheWikipedia pageof
Udemy
1
Set Easter and Christmas aseventsand Black
Friday asRegressor
2
Visualizeresults3
Measureaccuracy4
Test Set mustbe31 days5
Tags