jean-eric

Strange result with a simple 2-layers NN

3

Hello, I will give a snippet ```python import jax import jax.numpy as jnp from jax import jit from jax import grad from jax.example_libraries import optimizers from jax.config import config config.update("jax_enable_x64",...

Question: which gradient_descent_mse/gradient_descent_mse_ensemble to be used for "finite width-infinite time" & "infinite width-infinite time"

2

Hi, Let define `d` the number of features, `N` a neurone numbers and `ns` the number of training samples eg. `d:20, ns:200, N:200`. Here is a simple 2-layers Dense NN...

Question: difference between kernel_fn from stax.serial and nt.empirical_ntk_fn(apply_fn)?

2

Hello, I'm a newby with your library which is really looks nice indeed and I would like to take benefit of it to make some exercices to illustrate private lecture...

ScipyBoundedMinimize : recompilation of function to minimize at each instantiation?

2

### Discussed in https://github.com/google/jaxopt/discussions/220 Originally posted by **jecampagne** May 9, 2022 Hello, Here a simple snippet: ```python import jax import jax.numpy as jnp from jax import vmap, jit import numpy...

ScipyBoundedMinimize & jit => error

2

Hello, Below I give a snippet that crashes when jit is activated on a function that use ScipyBoundedMinimize I am using jax 0.3.5 jaxopt 0.3.1 Here is the snippet ```python...

API convergence SWT/DWT

2

Hello, Is there a chance that DWT (wavedec) and SWT (swt) would share for instance the trim_approx argument to get the same output? C urrently `wavedec` returns `[cAn, cDn, ...,...

Hello, I'm running yarn test on Mac Catalina > FAIL src/math/operations/__tests__/prismOps.test.ts ● Test suite failed to run TypeError: groups.flatMap is not a function 4 | 5 | const groups =...

Indexing 2D (Request)

9

Hi, When using for instance ``` J, L= 3,4 S = Scattering2D(J=J,shape=(64,64),L=L,backend='torch', max_order=2) ``` I have the coefficient in a array with typically such dimension [Cin,25,16,16]. Now, if I use...

Very nice job! One little extension concerns the number of channels, ie. RGB=3 but it can nicely be genralized > def __init__(self, num_layers=5, num_features=64, **num_channels=1**): then the 1st conv_layer >...

rational_quadratic_spline

1

Hello, Sorry, I probably ask a simple question: - is there a way to understand how the width/height parameters enter in the x_pos/y_pos of the knots. - is there a...