Issue when including continuous covariate in PyTwoWay about pytwoway HOT 3 OPEN

Hi Jakob,

Thank you for using PyTwoWay and for reaching out!

The issue you reference is with the preconditioner, not the solver. The solver is actually iterative, so it doesn't check if the matrix is full rank. I'm planning to add an option to check if the matrix is full rank over the summer.

I recommend trying out the different preconditioner options to find one that works. You should specify 'preconditioner': [preconditioner] in your FE control parameters.

The preconditioner options are:

• None
• 'jacobi'
• 'vcycle'
• 'ichol'
• 'ilu'

The default is 'ichol', which is the incomplete Cholesky decomposition, which seems not to work for your data. I would recommend trying 'jacobi' next. You could always try disabling the preconditioner with None, although this could be slow. I'm not sure how well 'ilu' and 'vcycle' will work, but feel free to try them.

In addition, you can always try out different solvers to see which is the fastest. In case you want to avoid preconditioners entirely, setting 'solver': 'amg' in your FE control parameters will use the AMG solver which doesn't use a preconditioner. This is also Professor Lamadon's preferred solver for very large datasets.

Please let me know if this doesn't resolve your issue, and also please feel free to reach out again if you have any more questions or issues with the code!

Best,
Adam

from pytwoway.

Thank you for your response!

I tried to use solver “‘amg”, but I received this error:
_pickle.PicklingError: Can't pickle <class 'pyamg.multilevel.coarse_grid_solver..GenericSolver'>:

However, this object is not found as “pyamg.multilevel.coarse_grid_solver..GenericSolver”.

I’m very sorry to bother, but I am just not sure which pickle is missing. Do you have an idea what causes this issue?

Best,

Jakob

from pytwoway.

Hi Jakob,

That sounds like an issue with multiprocessing. I think there are two good solutions.

First, you could install the multiprocess package so that it uses dill instead of pickle.

Alternatively, you can run the code with 'ncore': 1. If speed is an issue then you probably want multiprocessing though.

Please let me know if this resolves the issue!

Best,
Adam

from pytwoway.