CapeCod exposure on-level vector

[jbogaardt]

CapeCodneeds to support an exposure on-level (or trend) vector to make it useful in the real world.

[jbogaardt]

Option 1 for API:

cl.CapeCod(on_level=on_level_vector).fit(X,sample_weight=sample_weight)

Pros: matches closer to sklearn api
Cons: puts "data" in as an "assumption".

Option 2 for API:

cl.CapeCod().fit(X,sample_weight=sample_weight, on_level=on_level)

Pros: puts "data" where it belongs
Cons: departure from sklearn

[jbogaardt]

Option 3 (Preferred):
Create a ParallelogramOLF transformer to transform sample_weight that creates an olf_ attribute on a Triangle. If sample_weight has the olf_ attribute, then CapeCod will use it.

# Assuming we have a rate history DataFrame with columns 'rate_change' and 'eff_date'
olf = cl.ParallelogramOLF(rate_history, change_col='rate_change', date_col='eff_date')
sample_weight = olf.fit_transform(sample_weight)
assert hasattr(sample_weight, 'olf_')  # sample_weight should have 'olf_' property
cl.CapeCod().fit(X,sample_weight=sample_weight) # cape cod will access 'olf_' property if available

[jbogaardt]

Refer to this example on how to use.