Margaret Duff

Results 90 comments of Margaret Duff

In gradient descent, we have: - Deprecated the armijo rule parameters alpha and beta - Deprecated `objective_function` in favour of `f` - Deprecated `atol` and `rtol`

CGLS: deprecate tolerance

There are a few different options to fix this: - Increase the default max_iterations in the Armijo rule - Change the error message, to include instructions for defining the step...

For an example of this, see CIL Demos 01_optimisation_gd_fista.ipynb and change this step size to None, the default ![Image](https://github.com/user-attachments/assets/0258dcef-3850-4b95-a02b-d7a1c033bf57)

2f23129 is to check https://github.com/TomographicImaging/CIL/pull/1844

I will push the merge on the CIL side to fix the CIL issue. Then we can update the CIL tag on the super build

Discussed in a CIL dev meeting this morning. Possibly needs a discussion about a parent class for first-order gradient descent methods: GD, ISTA/FISTA, APGD, to prevent duplication of this code

> Out of curiosity: How many GPUs are you running this on? 2 GPUs