Edoardo Balzani

Additionally, if one creates a 2 layer pytrees of data (layer1: variables to convolve, layer2: trials) and of `eval_basis` , layer2: trials) to be used in the convolution (this doesn't...

Some additional thoughts: score for scikit-learn has to return a scalar; this may create problems if we want to regularize the model (it is basically mandatory with real data) and...

I am less sure we need the `scikit-learn` dependency, `BaseEstimator` is not a generic base class, it has a lot of methods for validating the input and other methods that...

this is an example of what get_tags returns on Poisson regressor ``` model = lin.PoissonRegressor() model._get_tags() Out[6]: {'array_api_support': False, 'non_deterministic': False, 'requires_positive_X': False, 'requires_positive_y': True, 'X_types': ['2darray'], 'poor_score': False, 'no_validation':...

We should differentiate the solver from the regularizer. 1. Have GLM requiring: - A **regularizer** class: instantiate a solver abstracting away the different arguments required by proximal solvers vs others....

Another idea is to use pytrees, so that each neuron could have its own tree of parameters; this is more flexible and more in line with jax; we could use...

If we change to pytree we should modify the proximal operator for group lasso; the mask could be done using tree_map, which will simplify the code.

I add a snippet of code that is helpful for computing operation on trees with different dept: ```python import jax x = { "outer_1": { "inner_1": {"stim": jax.numpy.array([0]), "coupl": jax.numpy.array([1])},...

I thought of a way to reduce a deep tree to a sub-tree with given structure, I don't want to forget this, so I am posting here. The issue is...

Hi everyone! Thanks for the great library! I tried to produce the behavior that @gviejo was talking about with pygfx. I ended up with something that seems to work okish;...