Issues with forecasting

[kymiller17]

In the forecasting model with a certain data set I have occasionally encountered a situation where the forecasting did not create a confidence Interval in certain situations for the first year. Why does this happen? and is there anyway to solve this? 

[kymiller17]

Forecasting Issues

 (I posted this in a seperate thread earlier but we figured I should post it here as well).

With the forecasting function in powerbi I have occasionally encountered an issue with a certain data set where the forecast did not produce a confidence interval for the first year.  It happens at a seemingly random rate throughout that data set I havent noticed any trends and its not just data sets out of a certain subsection, but more often that not it does succesfully produce it.  

Here are some examples

The First picture is an example of when it didnt succesfully forecast, the second one is an example of when it did.  I've attached the excel forecasting info as well.  The third and fourth are both examples of when it didnt but with just the data provided by excel vs powerbi.

If anyone knows anything as to why this may be happening, and how to fix it it would be greatly appreciated.

Thanks so much,

Kyle Miller

[kymiller17]

With the forecasting function in powerbi I have occasionally encountered an issue with a certain data set where the forecast did not produce a confidence interval for the first year.  It happens at a seemingly random rate throughout that data set I havent noticed any trends and its not just data sets out of a certain subsection, but more often that not it does succesfully produce it.  

Here are some examples

The First picture is an example of when it didnt succesfully forecast, the second one is an example of when it did.  I've attached the excel forecasting info as well.  The third and fourth are both examples of when it didnt but with just the data provided by excel vs powerbi.

If anyone knows anything as to why this may be happening, and how to fix it it would be greatly appreciated.

Thanks so much,

Kyle Miller

[Anonymous]

Here is a link to the actual data in some of examples of this problem (as well as some instances where, filtering or unfiltering the data does not result in a problem):

https://drive.google.com/open?id=1MVJJaWV4jjVMBTZ2iAhFuWchEjROTbsDUemtOLFHhPM

[Anonymous]

@v-yuezhe-msft Please see above post for the data you requested, and feel free to combine our two threads! We accidentally posted the same topic at http://community.powerbi.com/t5/Desktop/Forecasting-Issues/m-p/319536

Based on what I read at https://powerbi.microsoft.com/en-us/blog/describing-the-forecasting-models-in-power-view/ I suspect that the model used for the forecast is producing an error only when the data falls in a specific type of pattern. In other cases, the function it is applying does not have a problem. In our problematic cases, it must be choosing a particular forecast formula-of-best-fit that has a bug in it.

[Anonymous]

I have just now opened a support ticket to address this.

[Anonymous]

I opened a support ticket to address this.

[v-yuezhe-msft]

@kymiller17,

I am unable to view the image you post, could you please upload it again? In your scenario, do you set Confidence interval to 99%?

I test the forecasting feature in the latest version of Power BI Desktop(2.52.4921.682), everything works well. I will appreciate that if you can share sample data of your table, I will test it in my Desktop.



Regards,
Lydia

[Anonymous]

Hi Lydia - I'm not sure why the image isn't showing up for you, but here are some more. I have tried all confidence levels and other options. As you can see in these images, the problem moves from graph to graph when filters are applied. 

All of these visualizations are pulling values from the same data fields, but each graph is filtered to a particular segment. The value is just a count of a text field. Dates are dates. The data set is somewhat large and also confidential, and since the problem moves around it's not easy for me to share an example.

My colleague who posted above is looking into samples of the data and trying to reproduce the problem in Excel. He can post more info if he finds any conclusions.

Thanks for the reply.

no filterwith filter

[Anonymous]

Thank you for posting this for me, @kymiller17. Here is a screen shot of our problem. It seems verry unlikely that there would really be a confidence interval width of 1 on this data, and the other datasets this appears in (depending on the filters used, it's happening in multiple visualizations). Although we can theoretically circumvent the issue by setting the forecast to ignore the last data point, this works poorly with the filters, because the forecast then becomes unhelpful when specific filters are set.

Does anyone have any insight?