Hi,
Thank you for nice package. It's very useful.
But I have a question. How dose SciANN model calculate differentiation the constraint equations?
In practical, it needs two values to calculate differentiation.
Such as,
du/dt = (u2-u1) / (t2-t1)
I guess differentiation of constraint equations are calculated with tf.gradients in this SciANN package.
However, input and output of model are one-to-one correspondence.
Thus we have only one u and one t. How does model calculate differentiation?
I'm beginner of coding. So sorry if I say besides the point.
It is done through Automatic Differentiation -- you can think about it as analytical differentiation. Therefore, there is no need to do a finite difference type differentiation as you have up there.
Hi, Thank you for nice package. It's very useful. But I have a question. How dose SciANN model calculate differentiation the constraint equations?
In practical, it needs two values to calculate differentiation. Such as, du/dt = (u2-u1) / (t2-t1)
I guess differentiation of constraint equations are calculated with
tf.gradients
in this SciANN package. However, input and output of model are one-to-one correspondence. Thus we have only one u and one t. How does model calculate differentiation?I'm beginner of coding. So sorry if I say besides the point.