New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Adding Koopman operator evaluation, ITS evaluation and CK test to VAMPnet module #131
Comments
I'm sorry I hadn't realize the parameter of
So from this it seems that evaluating the assignment probabilities using softmax output distributions isn't compatible with estimating the Koopman operator using the VAMP estimator. From the reference
I understood that both can be considered correct at the same time: so you build the Koopman operator using the VAMP estimator and the assignments of each 'frame' to belong in one state are defined by the softmax output distributions.
How may I be able to know what is the probability assignment of each frame to a specific 'metastable state'? Thanks for you help |
The softmax output distributions can be understood as kind of a prior on your network (as in: you expect the data to possess a set of
You can find an example here: https://github.com/markovmodel/pyemma-workshop/blob/master/notebooks/08-vampnets-session.ipynb |
Thanks this is helpful, I do have a question regarding the notebook. In the 2D toy model paragraph the VAMPnet is used with 2 output nodes, meaning that the output states would be 2, which is exactly the dimension of
now I would expect that this would present a 2x2 koopman operator since the number of output states of the network was 2 but here it has only 1 dimension, is there something that is being ignored? I've tried also to put the |
In the 2D toy model the VAMPNet has only one output node which captures all of the variance. |
So in that case your Koopman matrix is a |
First thanks again for the very quick answer.
and the fit is called as follow Removing the |
Ah yes, now I understand what you mean. In this two-state probability distribution everything is already captured in the first component, the second is just |
oh I see it now! so the VAMP model from a model using Softmax is always D-1 dimensions since the last dimension would be always 1-other_components. Although this may be obvious it gets a little confusing if it's not explicitly said. Thank you very much! |
Yes that is right 🙂 as a final note, you may also get higher-dimensional VAMP models if you have more than two output states in the VAMPNet. By no means you always end up with a one-dimensional model (would be kind of cool though 😀 ) |
I misread, thought you wrote |
I contacted you a few days ago regarding the VAMPnet usage. I solved everything but I think there's something missing. From the references it is reported that you could evaluate the ITS and the CKtest straight from the koopman matrix built on the transformed data from your VAMPnet. Starting from these data it isn't difficult to evaluate by your own the koopman operator, the ITS and the Chapmann-Kolmogorov test but for completeness I would ask these functions to be added so that the users could entrust your implementation, for sure much more robust and compact.
Again congratulations for this huge and extremely well organized library!
The text was updated successfully, but these errors were encountered: