1
$\begingroup$

Most of the traditional machine learning algorithms need a feature vector of a constant dimension to predict the label.

Which algorithms can be used to predict a class label with a shorter or partial feature vector?

For example, consider a search engine. In search engines, when the user types a few letters, the search engine predicts the context of the query and suggests more queries to user.

Similarly, how can I predict a class label with an incomplete feature vector? I know one way is to pad the sequence, but I want a better solution.

$\endgroup$
  • 1
    $\begingroup$ Something that works at character level? search engines are more complex though $\endgroup$ – SajanGohil Jun 5 at 12:06
  • $\begingroup$ can you please introduce me some paper to study and get the idea? $\endgroup$ – Atena Jun 6 at 13:03
  • 1
    $\begingroup$ I am not sure about any specific paper, but now I realize you want something like partial embedding? Make do with 20 dimension out of 50? But for the search engine example, you can predict the whole word or query based on characters already input (there are other algorithms/DS like 'tries' for word completion but I don't think that's what you are asking for). It would help if you give a more proper example or a description for a more specific thing. $\endgroup$ – SajanGohil Jun 6 at 13:59
  • $\begingroup$ Thanks dear @SajanGohil. You are right. One solution is to predict the query based on the typed characters and then give it to model. in my question, imagine i have a tabular data in which number of features is constant (e.g. n). I need to use a model which can process samples with fewer features without padding the shorter sample. $\endgroup$ – Atena Jun 7 at 14:22

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.