Upload of multiclass classification dataset

[VijayanR]

How to upload multiclass classified dataset into Thingworx analytics . When i try to import, it showing error and not allowing to import more than two class dataset. ( True/false )

[zyuan1]

We have properties in String type in a dataset, and these properties contain more than 2 values, and this won't affect the import to the Analytics server. 

For example: 

Property Name: CoolDownType

Value:  Water; Air; Oil    etc. 

What's your dataset look like?

[VijayanR]

We have explored your suggestion and  we found the following observation.

1. Data upload is possible for multiclass with Option(a) : String & categorical & Option (b): String & continuous
2. Model building is possible with only option(b)
3. Here problem is ThingWorx analytics converts the Classification problem to regression problem.
4. Sample dataset is given below (Few rows are mentioned )

#	Speed	Flow	Head	torque	faults
1	2707.4	14.778	44.66	16.045	broken_blade
2	2965.6	33.591	16.438	7.5257	cavitation
3	2874.8	39.92	9.7401	10.019	healthy

[zyuan1]

If you have the TWA Extension Package with you, and you'll be able to find the sample dataset like BeanPro and Parking.  The expected goal field for these 2 datasets are Numeric type, and we can use the Categorical properties as dependencies for prediction.

[cmorfin]

Hi @VijayanR 

I know this is starting to be an old thread but I wanted to bring some additional info here.

If you use categorical goal, those are not currently supported in ThingWorx Analytics Builder, so this is why you can't build a model if using Builder.

However you can use such goal and build a model by using the services from the TrainingThing itself.

The Tech Tips How to work with ordinal and categorical data in ThingWorx Analytics give some more information on this.

Hope this helps

Kind regards

Christophe