Apparently at some point we decided to be ok with cases in corpus.py to NOT pass in make test, i.e. not generate a solution given the parameters of the setup. Also, we set the parameter max_stack_depth=20 when running tests to speed them up. All this led to a whole bunch of tests silently failing without ever producing the desired resolution.
Here this is changed again.
All test cases in corpus.py must yield the desired resolution when running make test - else tests fail
To achieve this I made the following changes to the test runner
max_stack_depth = 100
Use the scoring model - not the DummyScorer; but only when running tests, NOT when generating training samples
Allow for the latter distinction
Still, some tests were failing. I had to add extra "weight" for YYYYMMDD - YYYYMMDD intervals and slightly changed
a very specific case to avoid needing an even higher max_stack_depth
Apparently at some point we decided to be ok with cases in
corpus.pyto NOT pass inmake test, i.e. not generate a solution given the parameters of the setup. Also, we set the parametermax_stack_depth=20when running tests to speed them up. All this led to a whole bunch of tests silently failing without ever producing the desired resolution.Here this is changed again.
corpus.pymust yield the desired resolution when runningmake test- else tests failmax_stack_depth