I have the following dataset. For every time point (at a frequency of 1 hour), we can construct a graph consisting of 20 nodes representing countries. Each country (node) is characterized by 5 variables. The edges between countries are directed and weighted. Now, let's assume we have one year's worth of data, equating to 24*365 unique graphs. Each graph is associated with a label that we aim to predict. How would you approach such problem?

  • $\begingroup$ A graph can be represented by a matrix. Each of your countries along the rows and columns, properties of each country can be represented as several vectors. So all in all you have a matrix+several vectors per each observation. That would be a faithful albeit, perhaps, non-optimal representation $\endgroup$
    – Cryo
    Commented May 9 at 20:55


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.