Open robertour opened 6 years ago
The wald_wolfowitz is, in principle, a two valued test.
The https://github.com/brentp/skidmarks/blob/master/skidmarks.py#L85 would count the runs for the value/digit/symbol that appears first to count the runs; and the https://github.com/brentp/skidmarks/blob/master/skidmarks.py#L86 would builds the second group with all the rest of the values/digits/symbols.
It is suggested by @storchi in the comments in this SO question that a generalization should be possible.
This seems correct, as the tests depends on the expected mean runs and the expected variance runs (https://github.com/brentp/skidmarks/blob/master/skidmarks.py#L88-L92)
Looking now for a reputable source of how to correctly implement this generalization.
@psinger You would do something like this: https://github.com/psinger/RunsTest
From the README.md:
Could you clarify which of the tests are OK for non-binary (multivariate) and continuous (real values)?
As discussed in https://github.com/brentp/skidmarks/issues/4, there is no restriction for autocorrelation, but what about the other three tests:
gap_test
,wald_wolfowitz
,serial_test
,I will probably have to research them and try to complete the documentation, but if you already know something, it would be a start.