Test functions needed

[msmdev]

We need test functions for GPCCA in general and especially for sorted_schur.
They should be placed in the respective sub folders in msmtools/tests and should seamlessly integrate into the unittest framework in MSMTools.

[msmdev]

@Marius1311 Still important point - any progress here?

[Marius1311]

Not yet, but we will get onto this very soon! We are aware this is crucial.

[msmdev]

@Marius1311 perfect, thx! :)

[michalk8]

Hi @msmdev ,
I'm finally starting to write the tests and was thinking that it would be great to port the GPCCA tests from the matlab repo into this one.
However, the tests are using some data from files which are not present in that repo (e.g. example_matrix_mu0.txt).
Could you please share with me these files?

[msmdev]

Hi @michalk8 ,
this matrix (as most other test data except the larger count matrix in count.txt) can be found at the end of unitTests.m and regressionTest.m, e.g. at https://github.com/msmdev/gpcca/blob/66b3cb079847328b01fe61361c952f8b1093f3a9/regressionTest.m#L2131 .
In principle it is a good start to port the tests of the MATLAB GPCCA.
Anyway, most of the tests are very specific to the explicit program structure and thus the amount of useful and portable tests reduces a lot. Also the test cases there are very simple and target the molecular/conformational dynamcs use case.

[Marius1311]

Re the count.txt matrix - @msmdev, you mentioned that this is not yes a transition matrix - how to we turn it into one?

[Marius1311]

Also, you mentioned that setupOnce(testCase) would be a good place to start - this could be important for you @michalk8.

[Marius1311]

I have noticed in my own tests already that there are issues with our python implementation when the matrix is reducible (e.g. absorbing states, see #6 ) - we should have a test for this and look into it. @msmdev, what should be the expected behaviour for reducible matrices?

[msmdev]

Re the count.txt matrix - @msmdev, you mentioned that this is not yes a transition matrix - how to we turn it into one?

Take a look here: https://github.com/msmdev/gpcca/blob/66b3cb079847328b01fe61361c952f8b1093f3a9/regressionTest.m#L2484

[msmdev]

Also, you mentioned that setupOnce(testCase) would be a good place to start - this could be important for you @michalk8.

This basically just sets up a test folder with the necessary files and contains/prepares all the known test data. It is the recommended setup mechanism for the matlab test environment.

[msmdev]

I have noticed in my own tests already that there are issues with our python implementation when the matrix is reducible (e.g. absorbing states, see #6 ) - we should have a test for this and look into it. @msmdev, what should be the expected behaviour for reducible matrices?

There shouldn't be problems in principle (at least from the side of theory) - GPCCA can handle this - it will just identify absorbing states as stable macrostates with just one member, if we allow this by setting n high enough. If n is to low (lower than the number of absorbing states), it will try to combine several absorbing states to a macrostate, I guess, and all other (non-absorbing) microstates might be altogether assigned into a single residue macrostate... We need to take a closer look here to find the technical reason for this issue...

[Marius1311]

okay, agreed.