I have a data set where devices are represented by a collection of variables. These variables consist of several properties like a name, datatype, driver, limit values, etc. (mixed data; quantitative and qualitative). The devices should be classified by knowing just a subset of these variables.

Example model of one device "Tank" represented by four variables:

  Variable |        Name        |      Datatype       |      Driver      
  1        | Tank1_Temperature  |        INT          |       S7
  2        | Tank1_FillingLevel |        UINT         |       S7
  3        | Tank1_ValveIn      |        BOOL         |       S7
  4        | Tank1_ValveOut     |        BOOL         |       S7

Example input subset of variables to be classified as "Tank":

  Variable |        Name        |      Datatype       |      Driver      
  1        | Tank1_ValveIn      |        BOOL         |       S7
  2        | Tank1_Temperature  |        INT          |       S7

Many machine learning algorithms take feature vectors as input. The problem I am facing is, that I do not only have a simple feature vector, but a collection of feature vectors. My questions are: Are there any methods or algorithms dealing with data structured like this? Or are there any similar problems where the solution could be adapted from?

One thing I came up with, is to simply concatenate the variables to a single line. The disadvantage of this approach is, that the order of the variables-subset to be classified can be permuted.

Please consider that a device can have several hundred variables. And the number of different device models is large.

Hopefully the concept of the problem is described comprehensible. If not, please let me know and i will try to clarify.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.