Poission L-shape example, how to import training data about deepxde HOT 6 CLOSED

In the line

data = dde.data.PDE(geom, pde, bc, num_domain=1200, num_boundary=120, num_test=1500)

you only need to specify the number of training points, and then DeepXDE will automatically set up the training points for you (random uniform in the domain).

If you don't want DeepXDE to do this for you, you can also specify the locations of the training points using:

dde.data.PDE(..., anchors=X)

where X is a N by d matrix of total N points, with each row is one point of d-dim.

from deepxde.

Hi,
Nive job !
1 - I manage to run this example but the simulation is very slow, it takes 15min, is it normal ?
2 - Is there a way to define multiple geometry domain ? I'd like to solve the heat equation in a decomposed domain \Omega_i (each one have a conductivity coefficient).
3 - When do you plan to switch to tensorflow 2 ?

Best regards.

from deepxde.

@hbadi

• Which version are you using, and which example? Update to the latest version if not.
• Yes, you can define a geometry composed from more than one sub-domains, e.g.,

geom1 = dde.geometry.Rectangle(...)
geom2 = dde.geometry.Disk(...)
geom = geom1 | geom2

For more details, check Fig 6 in the paper.

• Hopefully, TF 2 will be supported in one month.

from deepxde.

I ran the L-shape laplacian with deepxde installed from pip (version 0.5)

from deepxde.

There is a bug in v0.5.0. You can either upgrade to v0.5.1, or to the latest version 0.6.0. I recommend v0.6.0, but there is an API change in PDE class (check the examples).

from deepxde.

TF2 is supported.

from deepxde.