“How to import elevation data?”

[shilf123]

No description provided.

[praksharma]

Your post is too long to read. What is elevation data? Is it BC, IC or just additional observation data that you wanna integrate with PINN?

[reactingflow]

Hi, @praksharma

I'm new to the community and currently learning how to train PINNmodels using DeepXDE. I have some input-output data pairs obtained from simulations, and I intend to use a portion of this data for training the model and another set for testing it.

To integrate this data with my PINN, can I directly use these input-output pairs in training, or do I need to modify certain parts of the code to achieve this? So far, I haven't found relevant examples in the case library that could guide me on how to proceed. It seems that in examples where equation constraints are applied, the inputs to the neural network are typically generated through some method, rather than using external input-output data directly.

Could you please provide me with some guidance on how to effectively incorporate my own dataset into the training process of a PINN using DeepXDE? Specifically, I would like to know if there are any steps I should follow or modifications I should make to the existing codebase to ensure my data is correctly utilized for training and testing purposes.

Thank you in advance for your help!

Best regards,

[chironbang]

Hi @reactingflow,
You can take a look at this example: https://deepxde.readthedocs.io/en/latest/demos/function/dataset.html
In the code, they load the training and test data from a file.

Best regards,
Chiron

[reactingflow]

Hello @chironbang ,

Thank you for your response and the example you've pointed out. However, my situation is a bit different and I believe it necessitates a more detailed explanation. The data I have comprises only input-output pairs. This particularity seems to not be fully addressed in the example you mentioned.

Given this setup, integrating such data directly into DeepXDE for training purposes while ensuring that the physical constraints of the equations are respected poses some challenges. Specifically, I'm interested in understanding how to adapt the code to accommodate this kind of data, making sure that the model can learn effectively from these input-output examples, and correctly applying the trained model for predictions on unseen data.

please see: #487 (comment)

Best regards,