[python-package] Add tests for more create_tree_digraph() inputs

[daguirre11]

Contributes to #7031
BEFORE

AFTER

7 line difference in coverage for /lightgbm/plotting.py

NOTE

I know these are very specific tests, however, the random state is fixed making the model fixture deterministic. Given other tests depend on the model fixture I expect it to be in a stable state and not change anytime (at least soon)

[jameslamb]

Thanks!

Instead of creating a new fixture and sharing it among multiple tests, could you please just extend the existing test_create_tree_digraph() to be a "deep" test similar to the others in this file?

I don't believe that method modifies the Booster, it should be safe to run multiple calls over the same Booster in the same test.

And we don't use something like pytest-xdist to run tests in parallel, so it's ok to have a lot of testing happening in a single test case.

the random state is fixed making the model fixture deterministic.

I think here's you're referring to the breast_cancer_split fixture? Even with identical input data, there is still non-determinism in LightGBM (see #6731).

Please add the following parameters to the LGBMClassifier() call:

• deterministic=True
• n_jobs=1
• force_row_wise=True

[daguirre11]

Thanks!

Instead of creating a new fixture and sharing it among multiple tests, could you please just extend the existing test_create_tree_digraph() to be a "deep" test similar to the others in this file?

I don't believe that method modifies the Booster, it should be safe to run multiple calls over the same Booster in the same test.

And we don't use something like pytest-xdist to run tests in parallel, so it's ok to have a lot of testing happening in a single test case.

the random state is fixed making the model fixture deterministic.

I think here's you're referring to the breast_cancer_split fixture? Even with identical input data, there is still non-determinism in LightGBM (see #6731).

Please add the following parameters to the LGBMClassifier() call:

• deterministic=True
• n_jobs=1
• force_row_wise=True

Thanks @jameslamb I will make these changes.

@jameslamb Yes I was referring to the breast_cancer_split. I see now the lightGBM model is not without the deterministic argument.

[jameslamb]

Suggested change

	assert "count: 200" in graph_body
	assert "count: 200" in graph_body, graph_body

All of the CUDA jobs are failing on this case:

FAILED tests/python_package_test/test_plotting.py::test_create_tree_digraph - assert 'count: 200' in '\tgraph [nodesep=0.05 rankdir=LR ranksep=0.3]\n\tleaf0 [label=<leaf 0: 0.677> color=black fillcolor=white penw... \n \n \n > color=white shape=rectangle]\n'

(build link)

Let's add this error message to get a bit more logs about what's going on.

[jameslamb]

Triggered a run here: https://github.com/lightgbm-org/LightGBM/actions/runs/23394312312

[jameslamb]

That failed too. The output was like this:

FAILED tests/python_package_test/test_plotting.py::test_create_tree_digraph - AssertionError: 	graph [nodesep=0.05 rankdir=LR ranksep=0.3]
  	leaf0 [label=<leaf 0: <B>0.677</B>> color=black fillcolor=white penwidth=1 shape=ellipse style=""]
  	split1 -> leaf0 [label=yes color=black penwidth=1]
  	leaf2 [label=<leaf 2: <B>0.343</B>> color=black fillcolor=white penwidth=1 shape=ellipse style=""]
  	split1 -> leaf2 [label=no color=black penwidth=1]
  	split1 [label=<<B>Column_20</B> &#8804;<B>17.510</B><br/>count: 346> color=black fillcolor="#ffdddd" penwidth=1 shape=rectangle style=filled]
  	split0 -> split1 [label=yes color=black penwidth=1]
  	leaf1 [label=<leaf 1: <B>0.286</B>> color=black fillcolor=white penwidth=1 shape=ellipse style=""]
  	split0 -> leaf1 [label=no color=black penwidth=1]
  	split0 [label=<<B>Column_27</B> &#8804;<B>0.146</B><br/>count: 512> color=black fillcolor="#ddffdd" penwidth=1 shape=rectangle style=filled]
  	legend [label=<
              <TABLE BORDER="0" CELLBORDER="1" CELLSPACING="0" CELLPADDING="4">
               <TR>
                <TD COLSPAN="2"><B>Monotone constraints</B></TD>
               </TR>
               <TR>
                <TD>Increasing</TD>
                <TD BGCOLOR="#ddffdd"></TD>
               </TR>
               <TR>
                <TD>Decreasing</TD>
                <TD BGCOLOR="#ffdddd"></TD>
               </TR>
              </TABLE>
             > color=white shape=rectangle]