{"id":459,"date":"2014-06-23T10:00:58","date_gmt":"2014-06-23T10:00:58","guid":{"rendered":"http:\/\/kourentzes.com\/forecasting\/?p=459"},"modified":"2014-12-18T10:40:54","modified_gmt":"2014-12-18T10:40:54","slug":"intermittent-demand-forecasting-package-for-r","status":"publish","type":"post","link":"https:\/\/kourentzes.com\/forecasting\/2014\/06\/23\/intermittent-demand-forecasting-package-for-r\/","title":{"rendered":"Intermittent demand forecasting package for R"},"content":{"rendered":"

A new package for analysing and forecasting intermittent demand time series and slow moving items has been release for R. You can download the latest version from CRAN<\/a>.<\/p>\n

tsb: TSB (Teunter-Syntetos-Babai) method.<\/li>\n<\/ul>\n
Two demo time series are included to try the various functions: data and data2.<\/p>\n
To fit Croston’s to a time series, simply use the following command:<\/p>\n
> crost(data)<\/pre>\n
The function comes with a few different options. By default it optimises model parameters and initial values using the MAR<\/a> cost function. Using the flag cost<\/em> different cost functions are available. By default the smoothing parameter for the non-zero demand and demand intervals are optimised separately, but this can be controlled using the flag nop<\/em>. Similarly the initialisation values for demand and intervals can be optimised or not and set either manually, or to preset initialisation heuristics. These are controlled by the flags init.opt<\/em> and init<\/em>.<\/p>\n
Three different variants of the method are implemented:<\/p>\n
\n
\u00a0‘croston’, for the original method.<\/li>\n
‘sba’, for the Syntetos-Boylan approximation.<\/li>\n
‘sbj’, for the Shale-Boylan-Johnston correction.<\/li>\n<\/ul>\n
For example, to get SBA forecasts with optimal parameters for 12 months ahead and plot the results you can use:<\/p>\n
> crost(data,type='sba',h=12,outplot=TRUE)<\/pre>\n
This will provide the following visual output:
\n $\"tsID.fig1\"$ <\/a>
\nFunctions tsb<\/em> and crost.ma<\/em> allow similar level of control.<\/p>\n
The next interesting function in the package allows you to create simulated intermittent demand series. The simulator assumes that non-zero demand arrivals follow a bernoulli distribution and the non-zero demands a negative binomial distribution. For example to create 100 simulated time series, 60 observations each, use:<\/p>\n
> dataset <- simID(100,60,idi=1.15,cv2=0.3)<\/pre>\n
The two last inputs control the average demand interval and the squared coefficient of variation of the non-zero demand for the series.<\/p>\n
To get a better view of the generated series, or a real dataset, we can use the idclass<\/em> function. This categorises the intermittent demand time series according to existing classification schemes. In particular the following are implemented:<\/p>\n
\n
‘SBC’, classification proposed by Syntetos-Boylan-Croston.<\/li>\n
‘KH’, the revised classification by Kostenko-Hyndman using the exact separation.<\/li>\n
‘KHa’, as above but using the approximate separation.<\/li>\n
‘PK’, the further revised classification that deals with temporal aggregation by Petropoulos-Kourentzes<\/a>.<\/li>\n
‘PKa’, as above but using the approximate separation.<\/li>\n<\/ul>\n
Both ‘KH’ and ‘PK’ require the demand interval smoothing parameter of SBA as input. This can either by calculated automatically by the function, or given as an input using the a.in<\/em> flag.<\/p>\n
Finally there are two views for each classification. One summarised that merely provides the number of time series the class\/method or a detailed one that provides a scatterplot of the time series characteristics. Here are some examples:<\/p>\n
> idclass(t(dataset))<\/pre>\nThis gives by default the PKa classification.
\n $\"tsID.fig2\"$ <\/a><\/p>\n
> idclass(t(dataset),type='SBC')<\/pre>\n
$\"tsID.fig3\"$ <\/a><\/p>\n
> idclass(t(dataset),type='KHa')<\/pre>\n
$\"tsID.fig4\"$ <\/a><\/p>\n
> idclass(t(dataset),type='PK',outplot='detail')<\/pre>\n
$\"tsID.fig5\"$ <\/a>
\nNote that the PK and KH exact classifications, unless the smoothing parameters are provided, can be computationally expensive.<\/p>\n
More information about the various optimisation cost functions for these intermittent demand methods and the classification schemes can be found here<\/a> and here<\/a>. An interactive demo of various functions and optimisation options can be found here<\/a>.<\/p>\n
Related Posts<\/H3>
Intermittent demand & THieF – EJOR Editors\u2019 Choice Articles<\/a><\/li>\n
Automatic robust estimation for exponential smoothing: perspectives from statistics and machine learning<\/a><\/li>\n
Elucidate structure in intermittent demand time series<\/a><\/li>\n<\/ul><\/div>","protected":false},"excerpt":{"rendered":"
A new package for analysing and forecasting intermittent demand time series and slow moving items has been release for R. You can download the latest version from CRAN. The launch version contains the following functions: crost: Croston’s method and variants. crost.ma: Moving average with Croston’s method decomposition. idclass: Time series categorisation for intermittent demand. simID:\u2026 Read More »<\/a><\/span><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[10],"tags":[45,54,22,23,39],"_links":{"self":[{"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/posts\/459"}],"collection":[{"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/comments?post=459"}],"version-history":[{"count":0,"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/posts\/459\/revisions"}],"wp:attachment":[{"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/media?parent=459"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/categories?post=459"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kourentzes.com\/forecasting\/wp-json\/wp\/v2\/tags?post=459"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}