Time Series Insights (Preview): errors modelled as categorical variables slow and max. 100.

Occasional Contributor


A customer has a device which can generate errors. The IoT data is processed by Time Series Insights (Preview) in Azure.


I modeled the errors as an categorical variable. But I ran into some problems:

- the device has 370+ different errors and TSI has a maximum of 100 categories.

- the TSI Explorer becomes at times very slow

- the TSI Explorer does not show the original (error) value.


What is the best way to model a error value with more than 100 different labels in TSI?


Edit: TSI Explorer was very slow in Chrome, but it's ok in Microsoft Edge.


5 Replies

@mitchelh Unfortunately, 100 categories is our limit. 


I'm curious to learn what this statement means: TSI Explorer does not show the original (error) value. Could you explain further? 

@mitchelh additionally could you please let us know if those are error messages itself or error codes ! Its an interesting use case where a single device is emitting more 370 error codes/messages. If the device itself is programmable/configurable, is it possible to group them into different buckets (based on some criteria) and configure one telemetry channel per bucket ? 



When you hover above a categorical variable in the TSI Explorer you only see the label text (between brackets) but not the value. Maybe the value associated with the label could be included in the text. That would also be helpful in case of a value which has no category label associated with it, which then could display [Unknown (%d)]  with the value in the hover text.



The device is an ECU (Electronic Control Unit) of which we receive the error codes.
There could be some criteria to group some errors, but operator would want to known what specific error code it was.
@mitchelh, Sorry about the inconvenience. While we look for a permanent solution, I would like to recommend ( based on ECU’s ability to group error codes) to send the error codes via more than one channel. Perhaps one channel per group and error codes in each group doesn’t exceed 100!