data files for deeponet examples

[fperiago]

Dear users,

I cannot find the data files to run the examples included in deeponet. Specifically, does anyone know where can we download

antiderivative_aligned_train.npz

antiderivative_unaligned_train.npz

??

Thanks in advance, Best regards, Paco

[praksharma]

https://yaleedu-my.sharepoint.com/:f:/g/personal/lu_lu_yale_edu/EnTn0aLimaRJuNKDOc0lfHkB2MXK8n8vAO1oV5cWVdJo3w?e=OLp80r

the link is already provided in the docs.

[awecefil]

https://yaleedu-my.sharepoint.com/:f:/g/personal/lu_lu_yale_edu/EnTn0aLimaRJuNKDOc0lfHkB2MXK8n8vAO1oV5cWVdJo3w?e=OLp80r

the link is already provided in the docs.

Hi, could you explain this dataset? I have already read the paper of DeepONet, but still not clear about the data generation part

So far as I know:

1. In the antiderivative_aligned_train.npz dataset, there are three numpy array X_train0, X_train1, and y_train, and the shape is (10000, 100), (10000, 1), (10000, 1) which corresponds to (u, y, G(u)(y)) in the paper

What I want to know:

1. The relationship between u, y, and G(u)(y)
2. u seems to be a set of functions that are sampled from a Gaussian Random Field or Orthogonal Polynomials? But which one is correct for the shape of X_train0, (num_of_functions, num_of_points_for_each_function) or (num_of_points_for_each_function, num_of_functions)?
3. What is (10000, 1) means for X_train1? Can it be seemed as another 10000 points for u? If this is the case, why not y_train be like (10000, 100) or (10000, 10000)?
4. Follow Q3, I know that G(u)(y) is the label for supervised learning and is obtained from solving ODE or PDE by numerical method that mentioned in the paper. But what we get from the numerical methods? For example, y = [0.1, 0.5, 0.9, ...] and a function u(x) is sampled at x=[0, 1, 2, 3,...], so the numerical method does like interpolation to get u(0.1), u(0.5), u(0.9) as the value of G(u)(y)?