I am now searching for a long time on the internet and on papers for an answers of simple questions. Am I able to train a Support Vector Regression algorithm with different data sets? If yes, how is the approach called?

I have 10 times the same battery with different usage, temperature and capacity.

  • Usage and temperature are features (x_i,i) and capacity is the output (y_i,i).
  • Battery_1 till timepoint n: [x_1,1 y_1,1; ... ;x_1,n y_1,n]
  • ...
  • Battery_10 till timepoint n: [x_10,1 y_10,1; ... ;x_10,n y_10,n]

Now I want to train my SVR with these sets, where the samples within a set belong together. I want give the algorithm a set of usage and temperature where I don't know the capacity and my SVR should predict it as such : x_d --> y_d.

Thank you very much for your help and input.

  • $\begingroup$ Let me help understand your problem, you have 10 different datasets collected from same battery output and you want to feed these 10 data sets to a SVM model, correct? $\endgroup$ – Xformer Jul 31 '17 at 11:44
  • $\begingroup$ These are ten different batteries form the exactly same type. So every set belongs together. The "function" behind all sets is the same. The Inputs are different and so the output for every battery is also different from the other. $\endgroup$ – i.k. Jul 31 '17 at 11:46
  • $\begingroup$ Ok. as long as number of features you shouldn't be having problem them feeding to SVM. $\endgroup$ – Xformer Jul 31 '17 at 12:06
  • $\begingroup$ Thank you for the answer. Does then my training input matrix look like this: M=[x_1,1 ... x_1,n ... x_10,1 ... x_10,n]? (Analog the output matrix) Is there no loss of time information or loss of dependency within a set? $\endgroup$ – i.k. Jul 31 '17 at 14:12

Why not concatenating the 10 datasets you have and simply creating one big dataset? Alternatively you could train 10 seperate SVMs and let them vote (for examples: output = mean of the 10 SVMs).

  • $\begingroup$ How do I create a big dataset? Each series belongs together. How do I put them all togther without losing this information? Is it an option to create input-vectors with a additional feature of the output one timestep before? $\endgroup$ – i.k. Aug 8 '17 at 11:25

Add one more feature in dataset which will identify battery, with this your dataset will have 3 features battery,usage, temperature and one target variable capacity. Test this out if you are losing the information then create different models for each data set.

  • $\begingroup$ What would be happen if I create another feature with gives information about the time for this point? $\endgroup$ – i.k. Aug 2 '17 at 14:30

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.