Suryanarayanan Manoj Sanu

Hey guys, I was trying to use haiku to create a convolutional neural network architecture (to reproduce a paper, whose implementation was in TF2.0). The CNN works correctly, however, when...

second-order derivatives with implicit diff

2

I was trying to use the `custom_root` decorator to differentiate through a solver. When I try to take the gradients, it works well. However, if I try to use `jax.hessian`,...

Attempted boolean conversion of traced array - for hager-zhang

2

I was trying to use `jaxopt.LBFGS` with `linesearch='hager-zhang'`. The following error appeared ``` Attempted boolean conversion of traced array with shape bool[].. The error occurred while tracing the function _update_interval...

Failure to wait for the data_object creation

3

When I run a DOE with f3dasm, **sometimes**, a few nodes produce the following error and quit. ``` Error executing job with overrides: ['++hpc.jobid=4', 'hp_tune.model=baseline', 'hp_tune.model_seed=-1'] Traceback (most recent call...

``` rsampler = f3dasm.sampling.RandomUniformSampling(design, seed = 5) samples_random = rsampler.get_samples(2) ``` gives the result  If I run `samples_random = rsampler.get_samples(2)` again, it gives a different set of samples. ...

Using K-FAC with physics-based losses

3

Hey, Thank you for the implementation. From the guide, I saw that I have to register loss functions to be able to use K-FAC. For my specific case, the loss...

Memory leakage during optimization

2

It seems there is come memory leakage somewhere. It can be clearly seen if the topopt example is run and you monitor the memory. It is steadily increasing. This results...

- If we try to run a simple linear elastic problem with

In `solver.py`, the following lines ``` tol = 1e-6 while res_val > tol: dofs = linear_incremental_solver(problem, res_vec, A_fn, dofs, precond, use_petsc) res_vec, A_fn = newton_update_helper(dofs) # test_jacobi_precond(problem, jacobi_preconditioner(problem, dofs), A_fn)...

- At the moment, for some BCs (especially in TopOpt), the petsc version converges but the jax_solve does not. - From our tests, we understood that this is due to...