Why does the output of the function 'neural_net_vessel1' go to the exponential of A? about 1dbloodflowpinns

[hieronimoDos]

A is the area and it should always be non-negative. For this reason, we use a trick where we consider that the neural net prediction as the logarithm of the area and compute its exponential. So, this exponential you see here is used to cancel the logarithm. This trick, meaning tf.exp(A) where A is tf.log(A), always provides non-negative predictions.

[YouqiongLiu]

A is the area and it should always be non-negative. For this reason, we use a trick where we consider that the neural net prediction as the logarithm of the area and compute its exponential. So, this exponential you see here is used to cancel the logarithm. This trick, meaning tf.exp(A) where A is tf.log(A), always provides non-negative predictions.

What a good idea! Thanks.
For the prototype Y-shaped bifurcation, I cannot run the program 'Y_shaped.py' to reproduce the result as shown in Figure 6 in paper.
Is there any skill in here?

[hieronimoDos]

You need to run the code for a larger number of iterations. You need to change lines 103 and 104 in 'Y_shaped.py'

model.train(1,1e-3)
model.train(2,1e-4)

to

model.train(90000,1e-3)
model.train(40000,1e-4)

You can also check the paper to see the training details. I will correct the code to resolve any confusion.

[YouqiongLiu]

You need to run the code for a larger number of iterations. You need to change lines 103 and 104 in 'Y_shaped.py'

model.train(1,1e-3)
model.train(2,1e-4)

to

model.train(90000,1e-3)
model.train(40000,1e-4)

You can also check the paper to see the training details. I will correct the code to resolve any confusion.

Thanks. I've only trained 40,000 times before.

[hieronimoDos]

I will close this for now and if you need more help send me an email.