Jonathan Wenger
Jonathan Wenger
# Problem The test code checking the SuiteSparse matrix interface makes a call to https://sparse.tamu.edu/. This test can fail unexpectedly (typically with the below error message) if the website is...
Linear operators should adhere by similar broadcasting rules as ``RandomVariable``s and ``np.ndarray``s. For example ```python from probnum.linalg.linear_operators import Identity A = Identity(5) + 2 ``` should return a ``_SumLinearOperator``.
Currently random variables cannot be serialized using `pickle`, even when they are fully defined by their parameters, such as `Normal`. This feature is very useful when trying to save results...
# Idea The probabilistic linear solver can solve problems with multiple right hand sides `nrhs` by taking multiple actions at each iteration, one for each right hand side. The update...
Currently when initializing a normal random variable one cannot simply specify a Cholesky factor of the covariance but has to pass a covariance as well. ```python import numpy as np...
Currently the scale of the uncertainty calibration for probabilistic linear solvers isn't tested thoroughly enough. In particular the following tests could be added or expanded upon. - [ ] Uncertainty...
Currently tests (in particular for `probnum.linalg`) throw a lot of warnings. This is probably caused by early stopping. Ensure that warnings are caught within the tests.
- [x] Cleanly separate `_preprocess_linear_system` from `_init_solver` ~and make sure only one takes the assumptions on A, i.e. `assume_A` as an argument.~ - [ ] Create a common (abstract) class...
The notebook explaining random variables shown at https://probnum.readthedocs.io/en/latest/tutorials/probability.html can be expanded with how to index, mask, slice and reshape random variables. Functionality added in #134 . This is implemented by...