Hardcoded 1970 value in core/base.py

[genedan]

There's a hardcoded 1970 value in core/base.py:

chainladder-python/chainladder/core/base.py

Lines 142 to 145 in 6c9ce99

	if (
	pd.Series(development_date).dt.year.min()
	== pd.Series(development_date).dt.year.max()
	== 1970

I'm not sure what this is for, the meaning should become clear after this file goes through the annotation process. In any case, we've been moving values like this to the package __init__.py (from which I'd propose a distinct options module be spun-off eventually) to centralize the constants.

[henrydingliu]

1970 the unix origin. when development is in months, rather than dates, it gets treated as nanoseconds added to the unix origin by pandas. we probably should be refactoring. the check against 1970, or any static year, can cause issue.

for example, this snippet runs fine.

df = pd.DataFrame(
    {
        'origin': [1979, 1980],
        'development': [1980, 1980],
        'reported': [1.0, 2.0]
    }
)
print(cl.Triangle(
    data=df,
    origin='origin',
    development='development',
    columns='reported',
    cumulative=True
))

      1980-01
1979      1.0
1980      2.0

but this functionally identical snippet fails

df = pd.DataFrame(
    {
        'origin': [1969, 1970],
        'development': [1970, 1970],
        'reported': [1.0, 2.0]
    }
)
print(cl.Triangle(
    data=df,
    origin='origin',
    development='development',
    columns='reported',
    cumulative=True
))

a quick fix would be to migrate and enhance this check inside Base._to_datetime(). the long-term enhancement would be to allow accepting development months, and address @kennethshsu biggest pet peeve regarding valuation being named as development in the Triangle constructor.

i'm going to take off the great first contribution tag