-
-
Notifications
You must be signed in to change notification settings - Fork 41
Better documentation of input/output dimensions #76
Comments
(sorry, I'm not sure why the format is off) |
Or more directly, it should conform to what's described in the SciML Style for this: https://github.com/SciML/SciMLStyle#documentation . The new SciMLSensitivity.jl documentation is a good model to follow: https://sensitivity.sciml.ai/dev/ |
actually, I am also thinking about not to support the |
Thanks for looking into this! Some other remark regarding documentation: The docs of the "models" Looking at the code, however, it seems that both function define a fixed architecture. It is also interesting that they are almost the same - which is very surprising given the very different names! (There seems nothing especially Markovian in It may be more transparent to remove this wrapper functions from the API and instead define them in the application examples/tutorials. |
@scheidan As for the |
Now I understand your reasoning for the models, makes sense! :) A short sentence like: |
@scheidan Thanks for the advice. Please take a look at the PR. |
That a big improvement, thanks! Especially the models should be clear now. There are two points that are not well defined yet:
To avoid this confusion I proposed above that we first define
|
I can understand that this is not that straightforward for people who are not that familiar with neural operators. But I am not quite sure what I can do. The "dimension of data" here is literally the dimension of "data". Channel on the other hand is one of the dimensions of the "input". The reason why the "channel" appears in the "input" dimension is because one needs to concatenate the gride information to the "data". And this is also mentioned in the original paper with the definition of
So, should I need to copy that again in the doc? |
@scheidan Please take a look at the PR and see if this works for you 😄 |
I would like to merge and temporary close this issue. Please feel free to re-open the issue if needed. |
Thanks for this great package! I think the documentation could be improved be defining the expected input dimensions better.
I think some thing like this would help a lot:
Operator kernel layer
where
x \in R^d
,v_{t}\in R^c_in
andv_{t+1}\in R^c_out
.Arguments
The text was updated successfully, but these errors were encountered: