Johannes Brachem

> I had to disable the tutorials workflow for the time being. The problem is that, on the GitHub action runner, we get reticulate 1.2x, but we need 1.31 or...

Implement easier access to and manipulation of var inputs

8

This PR contains only a few lines of changed code, but I think it can drastically improve quality of life while building and manipulating Liesel models. ## Problem statement I...

Var constructors: new_param, new_obs, new_calc, new_value

6

For a while now, I have been a little unsatisfied with two things: 1. Incsonsistent UI: `Var` class vs. `param` and `obs` functions. See also #130 2. The status of...

Add splines functionality

4

This PR contains only the three most basic functions that enable Liesel-users to work with P-splines without leaving Python: - `equidistant_knots` - `basis_matrix` - `pspline_penalty` The functionality, in particular the...

Jax was pinned to solve this: #213

There are some modules that may fit into contrib: - `liesel.distributions` - `liesel.model.distreg` - `liesel.goose.optim`

What I think would be nice: - [ ] Add an informative repr that lets users know about the parameters covered by this kernel. - [ ] Give users the...

At the moment, Liesel lets you initialize a variable with a distribution without declaring it as a parameter or an observed variable. This can lead to inconsistencies in the model,...

Being able to turn off the call to `plt.show()` is helpful when using functions like `gs.plot_trace()` and logging to a remote server, for example using Weights and Biases.

Currently, the GraphBuilder can either convert all floats to `float32`, or not convert floats at all. However, we may also want functionality to convert all floats to `float64`.