TS Model Factory

[#1]

Estimate time series forecasting models for multiple groups at once using the autoregressive moving average (ARIMA) method or the exponential smoothing (ETS) method.

[GabrielCM]

This tool has a bug in newer versions.
I adjusted the script in R at line 337.

• After: fit_stat <- summary (this_model)
• Now: fit_stat <- accuracy(this_model)

and at line 626

• After: fit_stat <- summary (the.model)
• Now: fit_stat <- accuracy(the.model)

[mj89]

THANK YOU for finding this. I was wondering why this macro suddenly stopped working when I updated to 2022.3 version of the Designer and R tools. Doing as you described above changing those two lines fixed the issue.

For those getting a "WriteYXDBStreaming" error try doing this. 

[Watsa]

THANK YOU VERY MUCH! It works again. I just want to add a few step for a basic user like me.

1. Right click on the TS Model Factory tool. 

2. Open Macro

3. On R tool, update two lines posted above.

[james_doohan]

Sorry I am not very experienced with scripts. Can you provide an example of what to change? 

[Watsa]

Hi @jam1 

1. right click on the TS Model Factory tool and open the macro

2. update the script in R tool in the following lines. (I copied the whole code and paste into a text editor. Replace the value in line 337 and 626

3. Save. Go back to your workflow, the macro will be updated automatically.

Line 337.

• Before: fit_stat <- summary (this_model)
  After: fit_stat <- accuracy(this_model)

Line 626

• Before: fit_stat <- summary (this_model)
  After: fit_stat <- accuracy(this_model)

[james_doohan]

Thank you. It's still throwing an error so just verify:

Line 337 currently has: fit_stat <- print(summary(this_model))

I am replacing that text with: fit_stat <- accuracy(this_model)

Line 626 currently has: fit.stat <- print(summary(the.model))

I am replacing that text with: fit_stat <- accuracy(this_model)

[Watsa]

@james_doohan 
This is what I have on my script and it works. Mine does not have print(). The last screenshot below was the error before I changed the script. 

https://community.alteryx.com/t5/Alteryx-Designer-Discussions/SOLVED-TS-Model-Factory-Error-after-up...

[aesekody]

I am trying to do a time series forecast, and my ARIMA model is running the forecast from 2007 to 2100. my data spans from 2007 to 2023. I don't know what I am doing wrong

[mj89]

@aesekody

A few things I would check in case of an unexpected result that have helped me:

1. Make sure your incoming data's frequency matches your forecast frequency (i.e. for yearly forecast you want to roll your incoming data up to annual level before passing it through the forecasting macro, etc.). Additionally, make sure incoming data is sorted in chronological order. You may want to also use a TS Filler tool to fill in missing periods in your data with zeroes before passing data through this macro. 

2. Check the macro's settings to ensure all are correct. Time period type should match your data's frequency. Series starting point is a good idea to use; just make sure you have correct setting or will get a wrong result. This TS Model Factory's macro's object output (O) should be linked to the TS Forecast Factory's "M" input anchor. In TS Forecast Factory make sure desired number of future forecast periods is correct.

3. If you are still getting an unexpected result after trying above two points, put a Filter tool after the TS Forecast Factory's output to filter out the undesired periods. In the back-end this macro's models are trained on historical data; you may get some training data included in your forecast unless you filter.  I recommend writing the filter to be any period greater than where your current data ends if you are not interested in seeing training periods for the model. That will ensure only forecasts are displayed in the output. The training periods are useful when checking accuracy of the model itself.

I hope that helps!

[aesekody]

Thank you so much for your help. I retired after quite a few hours on this, and I believe the solution lies within the really good suggestions I have got on here. I will definitely feedback the results of this exercise. Thanks

[ndehghani]

Hi,

How can I download TS Forecast Factory as well?

[ndehghani]

The tool is so helpful, would you please help on how to use multiple grouping field? right now I am just able to group it based on just one field? 

[mj89]

@ndehghani you can download the TS Forecast Factory from this link: https://community.alteryx.com/t5/Community-Gallery/TS-Forecast-Factory/ta-p/878748

[Noemi1]

Hi, 
How can I download the actual tool?
I’m trying to download the TS Forecast Factory but am only getting a workflow. I have already installed the Predictive Tools in Alteryx Designer.

[mj89]

@Noemi1 both the TS Model Factory and TS Forecast Factory are technically macros (note the .yxmc extension rather than typical workflow which is .yxmd extension). They are basically their own "sub workflows" packaged as a tool. Once you download and install either macro you may have to search for it in the Designer application if it does not appear in the Time Series group automatically. In the current version of Designer you may need to establish in your user settings where your macros are stored. It is easiest to keep them all in one folder for Designer to find them. To do this, go into Designer and select Options --> User Settings --> Edit User Settings. Select the macros tab and click on the + icon. In the Search Path box select the (...) ellipses icon. Then navigate to where your macros are and click OK. Save your settings and then restart Designer. When you open Designer back up check the Time Series tool group and they should appear.  

You will need both macros in order to generate a resulting forecast. You can link the model object (O output) from TS Model Factory to the model input (M input) of TS Forecast Factory in order to pass the forecast model through the rest of your workflow and perform further manipulations if needed. You can also refer to the official help documentation at  https://help.alteryx.com/current/en/designer/tools/time-series/ts-forecast-factory-tool.html  if further assistance is needed. Best of luck!

[drewed206]

Is there a way to include features to the TS model?