maho3
commented
8 months ago Hi Jice,
Glad to hear that you are finding the package useful!
Yes, I tried something like this for the gravitational wave example, although in that we only had n_feature=1.
The 1D CNN embedding network we used can be found here . In your case, your inputs would already have a preset n_feature, so I would comment out the line x = x.unsqueeze(1) for your application. We put that in there to add an extra dimension to our inputs so they would work with the convolutions.
Then, you can see an example of its application in Section 3-2 of this notebook. Apologies that it is quite messy right now.
Let me know if that works or you run into any bugs with the further application!
Jice-Zeng
Hi all, The package is pretty flexible and powerful for posterior estimates of parameters. I am trying to learn the package. I took a look at the examples and jupter notebooks, I did not see the parameter estimation using multivariate time-series data, such as posterior P(theta|data) in which data has a format of (n_datasets, n_time step, n_feature). I just wonder do you have any examples to handle case with such time-series data? Thank you for the time and help.
Jice