Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Deep Dive into Facebook Prophet

Deep Dive into Facebook Prophet

A gentle introduction on Facebook Prophet model in R programming. This is a workshop session. You can see the notebook detail here https://github.com/mrboneclinkz/meetup_useR_Indonesia_fbprophet

Fiqry Revadiansyah

June 28, 2019
Tweet

More Decks by Fiqry Revadiansyah

Other Decks in Technology

Transcript

  1. Bukalapak Deep Dive Into Facebook Prophet Meetup Bukalapak X Komunitas

    R Indonesia X Machine Learning ID Bukalapak R&D Office, Bandung | 28-June-2019 Deep Dive into Facebook Prophet
  2. Fiqry Revadiansyah Data Scientist @Bukalapak Bukalapak Meetup user! Indonesia Experiences

    Data Scientist @Bukalapak – Shoping Experience Tribe, Squad Chat and New Users Technical Content Reviewer @Packt Publishing – Time Series Analytics Expert Author of GSTAR (Generalized Space Time Autoregressive) package in R Education BS, Statistics @Universitas Padjadjaran Research Interest: Time Series Analysis, Regression Modeling, Spatial Modeling, Machine Learning, Gamification Business Meetup Bukalapak X Komunitas R Indonesia X Machine Learning ID @Bukalapak R&D Office, Bandung | 28-June-2019 About me
  3. Session Info NO MATH/STAT FORMULA 01 ASK NOW OR NEVER

    02 PROACTIVE REWARD 03 No need to have a math/stat background to hype this topic! *Unless you ask Don’t hesitate, no hassle, just ask at your own moment! Ensure yourself fun, comfort, and proactively share your thoughts, then get the reward Bukalapak Deep Dive Into Facebook Prophet
  4. Rewards Bukalapak Deep Dive Into Facebook Prophet Just for 3

    best learners! *This e-book worth $26 (Amazon Kindle)
  5. Forecasting History Why Bukalapak Forecasting jaman old Forecasting jaman now

    From 250 – 900 AD until now, forecasting is still exist. It is mostly because forecasting technique require highest quality of art and human confidence Background: Time Series History
  6. 1970’ – 2000’ ARIMA (Box & Jenkins, 1970) ARCH/GARCH (Engle

    1982) State Space Model (Snyder, 1985) Artificial Neural Network (Zhang, 1998) Now, what is going on? 1950’ – 1970’ Exponential Smoothing (Brown, 1959) Holt Winter SES (Holt, Winter, 1960) Decomposition (Brown, 1963) 2000’ - Now Spatio-Temporal Time Series Nonlinear Time Series Machine Learning Time Series Facebook Prophet Model Bukalapak Background: Time Series History
  7. “Internet Cepat Peramalan Tepat? Buat apa?" Bukalapak - Menteri Time

    Series dan Forecasting (2018 – 2016 SM) Background II: Time Series in Real Practice
  8. Myths - Useless - Cannot fit into production - Math

    and Stats Geek, not linear to business matter Bukalapak Time Series Forecasting Myth vs Fact Facts - Performance Measurement - Maintainable and able to be deployed - Capable to solve business problem Background II: Time Series in Real Practice
  9. Bukalapak FB Prophet Model Facebook Prophet Model *Forecasting at Scale,

    Taylor, S.J., and Letham, B. (2017) Time Series Model Founded to make automation of data forecasting for coping with the discrepancy between domain-based knowledge and statistical knowledge A Brief Preview
  10. FB Prophet Baseline Decomposition-Based 01 Feature of Business Time Series

    02 Analyst-in-the-loop 03 FB team believes that a time- series data is containing various components, such as trend, seasonality, holiday, etc. Compatible to business day, such as feature deployment date, national day-off, etc. Automate the process of surface problems, visually inspect forecasts, modelling, and evaluation. Bukalapak FB Prophet Model
  11. Bukalapak FB Prophet Model Why should FB Prophet? No Math/Stat

    Background Required Easy to Customize Long lasted Forecast Cited by Many Journals
  12. Thank you Fiqry Revadiansyah Data Scientist @Bukalapak @fiqryr Bukalapak Meetup

    user! Indonesia Meetup Bukalapak X Komunitas R Indonesia X Machine Learning ID Bukalapak R&D Office, Bandung | 28-June-2019 mrboneclinkz