When runs, we meet some puzzled points

[darioizzo]

please open a discussion item on github. Check your English and point directly to the code lines you refer to. Use markup for formatting your question. You can also embed images and all ...

https://github.com/darioizzo/geodesyNets/discussions

dario

On Fri, 24 Feb 2023 at 01:30, Shanhong Liu @.***> wrote:

Question 1: How to select the batch_size in during training when making mask? Turn to the loss function, if involving batch_size=m, the loss function should be set as this way? [image: image] https://user-images.githubusercontent.com/24770360/221062399-7b2242f8-6e6b-4196-97bf-96662089a51c.png

Question 2: In the drawing module (def plot_model_rejection ()), we first generate candidate points with uniform distribution. Should we judge the candidate points whether within the model or not? Then we used the candidate density and the mask together, but here what is the purpose of the mask here?

rho_candidates = model(nn_inputs).detach() c mask = rho_candidates > (torch.rand(batch_size , 1) + crop_p)*

What is the role of c in this case? And then plus a density minimum value (cross _ p), how to get this cross_p value?

Question 3: In the drawing module (def plot_model_rejection), it uses a judgment that the candidate point is empty or not (screenshot below:)

if len (candidates) == 0: print ("All points rejected!Plot is empty, try cropping less?") return

What is it meaning?

Question 4: In the process of differentiation training, because the density can be a negative value, the mask may lead to the model rejection in the drawing part? My current results for Differential Training are shown below [image: image] https://user-images.githubusercontent.com/24770360/221062355-67b9be23-415f-4acd-a72e-d9d8dc829734.png [image: image] https://user-images.githubusercontent.com/24770360/221062366-dd2a3692-5df6-456f-b6a0-4b07115a1c30.png

— Reply to this email directly, view it on GitHub https://github.com/darioizzo/geodesyNets/issues/107, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAZMI34TGX3QFIAQBQJ6WZDWY76L7ANCNFSM6AAAAAAVGJEAHQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>