Right now growthfactor.py incorrectly uses factor to mean "rate" (whereby in TEPs-I it's a multiplicative factor). This has been corrected the countmatch_matcher branch, but we should clean up some of the outstanding issues, like the existence of a base year.
More generally, however, we should also consider point 2 in #25 - should we instead store the year-on-year growth rates (calculating growth rates for each year separately from other years), then allow subsequent methods to multiply these growth rates together to get prod(GR_k)?
Tasks:
[ ] Discuss a more reasonable growth factor calculation with Aakash. If we calculate growth only from a single year (multiplying growth factors from multiple years together), does that increase the noise? Should we be performing exponential fits on weekly or MADT data rather than just doing a two-point AADT fit?
[ ] Remove redundant attributes in PermCount, like base_year.
[ ] Swap from the multi-year fitting strategy to a single year fitting one, storing the growth factors from each year so we can multiply them together for multi-year growth.
[ ] Simplify the exponential fitting algorithm to a two-point AADT fit ln(AADT_1 / AADT_0) / (year_1 - year_0) like Arman uses.
Right now
growthfactor.pyincorrectly usesfactorto mean "rate" (whereby in TEPs-I it's a multiplicative factor). This has been corrected thecountmatch_matcherbranch, but we should clean up some of the outstanding issues, like the existence of a base year.More generally, however, we should also consider point 2 in #25 - should we instead store the year-on-year growth rates (calculating growth rates for each year separately from other years), then allow subsequent methods to multiply these growth rates together to get
prod(GR_k)?Tasks:
PermCount, likebase_year.ln(AADT_1 / AADT_0) / (year_1 - year_0)like Arman uses.