fixes grain error in learning.py

[danielfong-act]

There is a small grain error in learning.py that affects semiannual grain triangles. As it stands, passing a semiannual triangle to learning.py (via some estimator) causes a size mismatch error during triangle arithmetic.

here's an example largely ripped from an earlier grain issue (#609)

import chainladder as cl, pandas as pd
# Create semi-annual triangle data
data = {
    'origin': ["2007-01-01", "2007-01-01", "2007-01-01", "2007-01-01", "2007-01-01", "2007-01-01", "2007-01-01",
               "2007-07-01", "2007-07-01", "2007-07-01", "2007-07-01", "2007-07-01", "2007-07-01",
               "2008-01-01", "2008-01-01", "2008-01-01", "2008-01-01", "2008-01-01",
               "2008-07-01", "2008-07-01", "2008-07-01", "2008-07-01",
               "2009-01-01", "2009-01-01", "2009-01-01",
               "2009-07-01", "2009-07-01",
               "2010-01-01"],
    'development': ["2007-01-01", "2007-07-01", "2008-01-01", "2008-07-01", "2009-01-01", "2009-07-01", "2010-01-01",
                    "2007-07-01", "2008-01-01", "2008-07-01", "2009-01-01", "2009-07-01", "2010-01-01",
                    "2008-01-01", "2008-07-01", "2009-01-01", "2009-07-01", "2010-01-01",
                    "2008-07-01", "2009-01-01", "2009-07-01", "2010-01-01",
                    "2009-01-01", "2009-07-01", "2010-01-01",
                    "2009-07-01", "2010-01-01",
                    "2010-01-01"],
    'loss': [100, 200, 300, 400, 500, 600, 700, 
             150, 300, 450, 500, 550, 600, 
             200, 250, 350, 400, 450, 
             50, 100, 150, 200,
             100, 200, 300,
             50, 150, 
             100]
}

df = pd.DataFrame(data)

tri = cl.Triangle(
    df, 
    origin='origin',
    development='development',
    columns='loss',
    cumulative=True
)
# ODP fit for example
BZsemi = cl.BarnettZehnwirth(formula = 'C(development)+C(origin)').fit(tri)

ValueError                                Traceback (most recent call last)
Cell In[1], line 36
     27 df = pd.DataFrame(data)
     29 tri = cl.Triangle(
     30     df,
     31     origin='origin',
   (...)     34     cumulative=True
     35 )
---> 36 BZsemi = cl.BarnettZehnwirth(formula = 'C(development)+C(origin)').fit(tri)

File ~\chainladder-python\chainladder\development\barnzehn.py:54, in BarnettZehnwirth.fit(self, X, y, sample_weight)
     49     warnings.warn("Using more than one linear predictor with BarnettZehnwirth may lead to issues with multicollinearity.")
     50 self.model_ = DevelopmentML(Pipeline(steps=[
     51     ('design_matrix', PatsyFormula(self.formula)),
     52     ('model', LinearRegression(fit_intercept=False))]),
     53             y_ml=response, fit_incrementals=False).fit(tri)
---> 54 resid = tri - self.model_.triangle_ml_[
     55     self.model_.triangle_ml_.valuation <= tri.valuation_date]
     56 self.mse_resid_ = (resid**2).sum(0).sum(1).sum(2).sum() / (
     57     np.nansum(tri.nan_triangle) -
     58     len(self.model_.estimator_ml.named_steps.model.coef_))
     59 self.std_residuals_ = (resid / np.sqrt(self.mse_resid_))

File ~\chainladder-python\chainladder\core\dunders.py:270, in TriangleDunders.__sub__(self, other)
    269 def __sub__(self, other):
--> 270     obj, other = self._validate_arithmetic(other)
    271     if isinstance(obj, TriangleGroupBy):
    272         def f(k, self, obj, other):

File ~\chainladder-python\chainladder\core\dunders.py:35, in TriangleDunders._validate_arithmetic(self, other)
     33 """ Common functionality BEFORE arithmetic operations """
     34 if isinstance(other, TriangleDunders):
---> 35     obj, other = self._compatibility_check(self, other)
     36     if obj.is_pattern != other.is_pattern:
     37         obj.is_pattern = other.is_pattern = False

File ~\chainladder-python\chainladder\core\dunders.py:69, in TriangleDunders._compatibility_check(self, x, y)
     63 x, y = set_common_backend([x, y])
     64 if (
     65     x.origin_grain != y.origin_grain
     66     or (x.development_grain != y.development_grain and
     68 ):
---> 69     raise ValueError(
     70         "Triangle arithmetic requires both triangles to be the same grain."
     71     )
     72 return x, y

ValueError: Triangle arithmetic requires both triangles to be the same grain.

This is caused by the wrong number of periods per year being given for semiannual triangles: there are 2 semis in an annual, not 6. This leads to self.model_.triangle_ml_ being constructed with a finer granularity than it ought to have. I fixed the incorrect values and reordered the grain dictionaries to make things more obvious. The above code runs as expected with this change.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.44%. Comparing base (358c89c) to head (f2d61bf).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #633   +/-   ##
=======================================
  Coverage   84.44%   84.44%           
=======================================
  Files          84       84           
  Lines        4823     4823           
  Branches      610      610           
=======================================
  Hits         4073     4073           
  Misses        538      538           
  Partials      212      212           

Flag	Coverage Δ
unittests	84.44% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[kennethshsu]

Good eye, can you add some tests too? :)

[danielfong-act]

@kennethshsu sure, were you looking for grain tests just for learning.py?

[kennethshsu]

Yes, I think that'll be good if you are up for it!