ARIMA and ETS Prediction

[jmarleigh]

Hi,

I just a quick question about the Predictive tools within Alteryx. I was wondering if it is possible, if you already have existing time series data, to allow Alteryx to predict beyond the data that is already in place. I've seen situations where users have given Alteryx specific data to allow it to predict from scratch what the outcome will be using either the ARIMA or ETS method, but nothing about doing what I suggested. It seems like it should be possible going from the ETS or ARIMA tool into the TS Plot tool, but so far I haven't successfully been able to make that happen.

Any help would be much appreciated.

Thanks,

Justin

[JohnJPS]

Hi @jmarleigh

I believe you'll need to use a TS Forecast tool to make the future predictions.  I can't see any other way.

[jmarleigh]

@JohnJPS - I just got there in regards to feeding ARIMA into the TS Forecast tool. One more question, is it possible to forecast for multiple items in a single workflow?

[JohnJPS]

Hi @jmarleigh

The easy way would be to run from one source into multiple ARIMA tools, and set each to forecast a different target field. If you wanted one tool to generate forecasts for multiple fields, you would have to write an iterative macro, which I'm guessing is certainly possible, but might get a bit involved.

[BridgetT · Accepted solution]

Hi @jmarleigh,

Actually, there are macros on the Predictive District of the Gallery that solve the problem of performing multiple predictions. The TS Model Factory macro creates models for multiple items at once, and the TS Forecast Factory macro estimates forecasts for these models. If you want more customization than these macros allow, @JohnJPS's approach will work. However, the R tool takes about 2 seconds to initialize every time, so putting an R tool in an iterative macro can create some performance concerns.

[MarySam]

Hi BridgeT,

I am using the TS Model Factory and TS Forecast Factory for co-variate Arima forecasting. The models worked well in the past, but recently I have been receiving this error in the TS Forecast Factory: 

Error: TS Forecast Factory (92): Tool #20: Error in forecast.Arima(mod.obj, h = nrow(covars), xreg = covars, level = conf.ints) :
Error: TS Forecast Factory (92): Tool #20: Execution halted
Error: TS Forecast Factory (92): Tool #20: The R.exe exit code (1) indicated an error.

Will you be able to shed any light on what they mean and how I can go about fixing it? Any help will be appreciated.

Thanks!

[Farhad]

Hi,

I am getting same error, just wondering, have you got the solution for this yet?

thanks,

Farhad

[jpburrowes]

For those that are getting the "forecast.Arima(mod.obj, h = nrow(covars), xreg = covars, level = conf.ints) :" error, the problem is due to including variables that aren't used as covariates in the covariate input into the Forecast Factory tool.

Only the covariate fields should be included in that input, so make sure to exclude (e.g.) the grouping variable and any other time-series variables that help to order the data. It'd be best to make sure the rows are sorted into the correct order first, so that the groups are in the same order as they are in the Model Factory, and the future values of the covariates are in the correct order, and then exclude all additional, non-covariate variables using a select tool. The columns of covariates going into the covariate input should also be in the same order as the selected covariates in the list in the Model Factory tool.

There is an example workflow available here, which has an example that uses one covariate: https://gallery.alteryx.com/#!app/TS-Factory-Sample/5772b0ebaa690a1348cc6bcb 

[JohnTheAnalyst]

@jpburrowes , I downloaded and looked at the example you gave for covariates above.  I noticed the input data set were forecasts for website hits (clearly not part of the original data set) so, I'm wondering; where did these forecasts come from?  

Thanks very much,