0
$\begingroup$

So I'm facing a problem where I have a sequence (30h of data with 10sec intervals) and which is labeled to a class (3 different classes) for the whole sequence. I'm used to work with time series who have an output for each time sample in the sequence and now I'm block with this.

So far I have thought of performing a statistical study of the sequence and create a dataset where each row contains the results of that study from a sequence and the target. This way I'll have a dataset with a Y for each X (but I'll lose a lot of information from the sequences)

I come here to see if any of you have faced a similar problem before or if you have any ideas of how to approach this.

$\endgroup$
  • $\begingroup$ Hello Jaime, What is the expected output of the network? What are you trying to achieve with the time series data? $\endgroup$ – Nischal Hp Nov 5 '18 at 16:04
  • $\begingroup$ Hey, expected output would be class classification (3 different classes). The time series data is just the data I've been given, my problem is that I don't know how to deal with the problem of having 30h of times series to be 1 class only. $\endgroup$ – Jaime Ferrando Huertas Nov 6 '18 at 8:12
1
$\begingroup$

With only one output, there is no task for a supervised learning algorithm for classification. Your algorithm can only predict your only one class, you do not give any other situations of other classes for your algorithm to work on; a classification algorithm will need at least two classes. I have strong experience with time-series learning problems, but it seems that your problem has nothing to do with whether your data is time-series.

However, depending on your task, you can implement unsupervised learning which does not require any labels. You can seperate your data into clusters which have similar properties as time-windows. Then you can predict with new (or test) samples that which cluster should it be from, what it is alike.

| improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Oh I was explaining the problem wrong. I do have different classes for a sequence of 30h data. $\endgroup$ – Jaime Ferrando Huertas Nov 6 '18 at 8:08

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.