One-Hot Encoding

[PErhams]

Hello,

I am currently working in ThingWorx Analytics and have categorical data as input. In order to try to enhance the accuracy it is common/recommended to try One-Hot Encoding(OHE) of the data. However, i don't think any of the opTypes are suitible for OHE, as this would require a vector as input. Has anyone tried to use OHE in ThingWorx analytics?

Best regards

[cmorfin]

Hi

I don't think this is something we can use but I need to look into this and will post once I have more details

Regards

Christophe

[PErhams]

That would be great! Since OHE is a farily common way to adapt categorical input it is probably something PTC should consider implementing. 

[cmorfin · Accepted solution]

Hi

Sorry for the delay, I was waiting for some answer from R&D on this.

It appears that one hot encoding is internally automatically applied to the field, though it depends on the model type.

This is not configurable, but used internally.

I have documented some more details about this at https://support.ptc.com/appserver/cs/view/solution.jsp?n=CS283787

Kind regards

Christophe

[PErhams]

Great, thanks!

Is there any documentation on pre-processing that is automatically done? Normalization/rescaling for example.

Regards,

Perham

[cmorfin]

Hi

There is no documentation on the internal functioning of ThingWorx Analytics.

Kind regards

Christophe