-
-
Notifications
You must be signed in to change notification settings - Fork 195
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
Adaptive reweighting or modify neural network to solve complex system. #708
Comments
You could modify the strategy you use to train the neural network to better fit the problem. This can be done like this:
This method allows you to select a weighting and number of samples for your problem. I found this to work for this specific problem, but you may have to experiment with some other weightings or samples for your problem to fit well. Please let me know if you have any questions about this and I will do my best to help you out |
I have tried with your code but I got the message
And got another message:
After using Pkg to add |
WeightedIntervalTraining is a new method of training that I developed. Can you verify that you have the latest version of NeuralPDE? There should be mentions of WeightedIntervalTraining in NeuralPDE.jl, ode_solve.jl and training_strategies.jl |
Oh, I forgot about that. Right now, I just use version 1.9.0. I will update the version and update the packages. |
I have updated the latest version, and your code is running well. For my code, I got a new error as follows:
My equation is:
I used Besides, I would like to ask that if my initial conditions are a large number (200 or 300), the network can work with it. because the sigmoid function has a limited value of the input (like it will work well in intervals from 0 to 1). If not, is there any way to normalize the input data? |
Can you post a full error, with the code that causes it? |
The error had been solved with the latest version of Julia and packages. |
I tried to use NeuralPDE to solve the ODE system. The results are not good when I used the code as a tutorial then I tried the adaptive_loss function and additional_loss function and got better results. My code is follow:
I got the results for the phase angle below:
However, when I solved the system by ODEPlroblem I got the results as follows:
We can see that the results from the two methods are approx in the first 5 seconds but after that, the neural network can not capture the nonlinear of the phase angle.
Please help me on how could I modify the neural network to capture the nonlinear functions for whole domains.
Thank you all.
The text was updated successfully, but these errors were encountered: