Exploratory Seminar: Time Series Forecasting with Prophet

Prophet is an easy to use time series forecasting algorithm developed by Sean Taylor and co. at Facebook. I’ll be demonstrating how to use it in Exploratory.

Transcript

1. EXPLORATORY
2. Kan Nishida co-founder/CEO Exploratory Summary Beginning of 2016, launched Exploratory, Inc. to make Data Science available for everyone. Prior to Exploratory, Kan was a development director at Oracle leading development teams for building various Data Science products in areas including Machine Learning, BI, Data Visualization, Mobile Analytics, Big Data, etc. While at Oracle, Kan also provided training and consulting services to help organizations transform by data. @KanAugust Instructor
3. Vision Make Data Science available for everyone
4. Data Science is not just for Engineers and Statisticians. Exploratory makes it possible for Everyone to do Data Science. The Third Wave
5. First Wave Second Wave Third Wave Proprietary Open Source UI & Programming Programming 2016 2000 1976 Monetization Commoditization Democratization Statisticians Data Scientists Smart Waves - Machine Learning / AI Algorithms Experience Tools Open Source UI & Automation Business Users Theme Users Exploratory
6. Questions Data Science Workflow Communication Data Access Data Wrangling Data Visualization Machine Learning / Statistics Exploration
7. Questions What you can do with Exploratory Communication Data Access Data Wrangling Visualization Machine Learning / Statistics Exploratory Data Analysis
8. Questions Communication Data Access Data Wrangling Visualization Machine Learning / Statistics Exploratory Data Analysis
9. Time Series Forecasting with Prophet
10. Time Series Forecasting with Prophet
11. • Application of Supervised Training • Build a model to predict value from date, with past data as training data. • By feeding the model with future dates, it will produce forecasted values for the future dates. Time Series Forecasting
12. Builds a model that predicts a day’s value from past N days. Traditional Approach with Time Series Forecasting Day 1 Day 2 Day 3 Day 4 Day 2 Day 3 Day 4 Day 5 Day 3 Day 4 Day 5 Day 6
13. • Time interval between data has to be same throughout the data • Day with NA is not allowed • Seasonality with multiple periods (Week and Year) is hard to handle • Parameter tuning by expert is necessary Problems with Traditional Time Series Model
14. • Open source Time Series Forecasting Algorithm from Facebook. • Designed for ease of use without expert knowledge on time series forecasting or statistics. Prophet
15. • Builds model by finding a best smooth line which can be represented as sum of the following components. • Overall growth trend • Yearly seasonality • Weekly seasonality • Holiday effects - X’mas, New Year, July 4th, etc. Time Series Forecasting by Prophet
16. Trend
17. Trend
18. Trend, Yearly
19. Trend and Yearly Combined
20. Adding Weekly
21. Trend, Yearly, and Weekly Combined
22. 22 Trend, Yearly, and Weekly Combined
23. 23 Holiday Effect
24. 24 Trend, Yearly, Weekly Combined, Holiday Effect
25. • Uneven time interval between data is not a problem. • Day with NA is not a problem. • Seasonality with multiple periods (Week and Year) is handled by default. • Works well by default setting. Parameters are easily interpretable Benefit of Prophet Approach
26. Time Series Forecasting by Analytics View Sales Data
27. Visualization by Line Chart
28. Select Time Series Forecast Analytics View
29. • Assign Order Date to Date/Time. • Select DAY (Floor to Day) as the unit of time. • Column, and Sales to Value Column. • Select SUM as the function. By default, forecasting period is 10. In this case, since DAY is the unit, forecast for 10 days will be made. Run Time Series Forecasting on Sales
30. Click Run button, and forecasted data is displayed in orange line. Let’s zoom into the period near the end
31. • Period with only orange line is the 10 days that is forecasted. • Area in light orange is uncertainty interval of the forecast. Zoomed into the period near the end of data.
32. Under Trend tab, actual values and trend line is displayed. Trend line is in green.
33. Under Yearly tab, yearly seasonal trend is displayed.
34. Under Weekly tab, weekly trend is displayed.
35. Under Data tab, table with forecasted values are shown.
36. • forecasted_value - The forecasted values • forecasted_value_high/forecasted_value_low - Uncertainty interval • trend - Overall growth trend • yearly - Yearly seasonal trend • weekly - Weekly trend Data with Forecasted Values
37. Advanced Topics
38. Holiday Effect
39. 39 • Holiday might effect the measures like sales, number of orders, etc. • Special events like conferences, black friday, sports events, etc. might effect the measures. Holiday Effect
40. 40 Unique Page Views of ”Apple Worldwide Developers Conference” (WWDC) at Wikipedia Data
41. 41 Apple Worldwide Developers Conference (WWDC) dates Data
42. 42 Data Future dates.
43. • Set ‘Day’ for Date / Time • Set 365 as Forecasting Time Period Forecast without Considering Holiday Effect 43
44. 44 These WWDC dates are not forecasted well by the model.
45. Add WWDC Dates by Join
46. Select ‘Full join’ as the join type because we want to bring the dates that don’t exist in the current data frame but does exist in the target data frame. Add WWDC Dates by Join 46
48. Select Holiday Column 48 These WWDC dates are not forecasted well by the model.
49. 49 Future conference dates are better forecasted by incorporating the holiday effect.
50. 50 Holidays Tab Green line is the Holiday effect.
52. Capacity
53. 53 Capacity - Upper / Lower Limit
54. 54 You can set Upper / Lower Limit values. • Sales / Demand ForecastingɿWe know it won’t exceed a certain limit. • Conversion RateɿIt never exceed 100%. Prophet can take those limits into account. Capacity
55. 55 You can set the lower limit when the upper limit is set. • The default value is 0. • You can set a negative value as well. Lower Limit
58. Trend Line 58 Trend line becomes a logistic curve instead of a linear line when Upper/Lower Limits are set.
59. 59 Evaluating Forecasting Model
60. • Keep the latest part of past data as test data • Forecast the test data period with the data before the test data period • Estimate the forecast Backtesting
61. 61 Assign Order Date and select ‘Floor to Month’ Assign ‘Sales’ to Value and select ‘Sum’.
62. 62 Switch to Test Mode Select ‘TRUE’ for Test Mode.
63. 63 Set ’12’, this means ’12 months’ because the data is currently set as monthly date. This will reserve the last 12 months data as the test data.
64. How to understand the result 64
65. Data is split into two sections. 65 Training Data Test Data
66. Since the forecasting model is built based on the training data, the forecasted values (Orange) and the actual values (Blue) are very close. 66
67. This is the data that the model didn’t know when it was built. 67
68. 68 The difference between the forecasted values and the actual values can be interpreted as how this year is different compared to the scenario if we assumed it could have been the same trend as the previous years.
69. 69 Compared to the previous years, we can see that there was a big drop in February and big increases in the later part of the year.
70. Model Summary 70
71. • MAE (Mean Absolute Error) : Mean of absolute differences between actual value and forecasted value. • RMSE (Root Mean Square Error) : Root of mean of squares of difference between actual value and forecasted value. • MAPE (Mean Absolute Percentage Error) : Mean of absolute differences in percentage of actual value. • MASE (Mean Absolute Scaled Error) : Divide MAE of the forecasting model by MAE of a naive forecasting model (a simple model that always forecasts one previous value). Metrics to Evaluate Time Series Forecast
72. The evaluation metrics are all calculated based on the difference between the actual and the forecasted values in this period. 72
73. Root
 Mean
 Square
 Error Square the Errors (Difference between the actual and the forecasted), take the mean, and root the mean. 73 RMSE (Root Mean Square Error)
74. 
 22 + 22 + 22 + 42 
 4 4 + 4 + 4 + 16 
 4 7 = 2.65 74 RMSE (Root Mean Square Error) 2 2 4 2 = =
75. Mean
 Absolute
 Error Absolute values of the Errors (Difference between the actual and the forecasted) and take the mean. 75 MAE (Mean Absolute Error)
76. 
 2 + 2 + 2 + 4 
 4 76 = 2.5 MAE (Mean Absolute Error) 2 2 4 2
77. Mean
 Absolute
 Percentage
 Error Absolute Percentage differences between the actual and the forecasted and take the mean. 77 MAPE (Mean Absolute Percentage Error)
78. 78 12 13 16 11 Take the actual values. MAPE (Mean Absolute Percentage Error)
79. 79 Take the errors. MAPE (Mean Absolute Percentage Error) 2 2 4 2
80. 80 Calculate the ratio of the errors against the actual values. If it’s negative make it positive. (Absolute) MAPE (Mean Absolute Percentage Error) 16.6% 15.4% 25% 18.2%
81. 
 16.6 + 15.4 + 18.2 + 25 
 4 81 Take the mean. = 18.8% MAPE (Mean Absolute Percentage Error) 16.6% 15.4% 25% 18.2%
82. Compare the Metrics by Market 82 To compare among Markets, select ‘Market’ for Repeat By and build multiple forecasting models.
84. 84 The drop in February is coming from Latin America (LATAM). February
85. 85 Unexpected increases in November are coming from Asia Pacific and Latin America. 11݄ 11݄
86. • RMSE and MAE indicate that the errors are smaller for Africa compared to Asia Pacific. • Can we conclude that the model for Africa is better? 86
87. Sales amounts are bigger in Asia Pacific than in Africa in general, hence it’s expected that RMSE and MAE are bigger for Asia Pacific. 87
88. • MAPE is calculated as the percentage of the errors compared to the actual values. • It is more useful when we want to compare the two without considering the scale of the actual values. • MAPE is smaller for Asia Pacific, which means we have a better model for Asia Pacific than for Africa. 88