Add meta_pin_num_card to proration grouping

[Damonamajor]

Quick fix to a long problem, to add meta_pin_num_card to grouping.

[Damonamajor]

@jeancochrane @wagnerlmichael
The thing that I've been noticing is that prorated pins are often multi-card. In our multi-card aggregating, I assume that we are summing sq_footage as though the entirety of that value is on one pin, where my understanding is that it should be shared between two pins. Is that a correct assumption?

[wagnerlmichael]

@jeancochrane @wagnerlmichael The thing that I've been noticing is that prorated pins are often multi-card. In our multi-card aggregating, I assume that we are summing sq_footage as though the entirety of that value is on one pin, where my understanding is that it should be shared between two pins. Is that a correct assumption?

I don't think we currently modify a prorated card's square footage contribution to the frankencard total sqft number. Perhaps it is worth testing that out, it makes sense to me intuitively. Could be worth consulting with valuations

[Damonamajor]

It seems to then use our 2 / 3 card technique below this, which I'm not sure is how we want it to interact with prorated pins.

[wrridgeway]

Is this supposed to be a draft still?

[Damonamajor]

Is this supposed to be a draft still?

yup, I want to check the model runs of 2025-10-05-didactic-rob and 2025-10-03-compassionate-gabe

[Damonamajor]

library(noctua)
library(DBI)
library(readr)
library(dplyr)

noctua_options(cache_size = 10)

con <- dbConnect(noctua::athena())

sql_query <- "
select meta_pin, meta_card_num, pred_card_initial_fmv
  from model.assessment_card
where run_id IN ('2025-10-05-didactic-rob', '2025-10-03-compassionate-gabe')
"

data <- dbGetQuery(conn = con, statement = sql_query)

test <- data %>%
  group_by(meta_pin, meta_card_num) %>%
  summarize(distinct_values = n_distinct(pred_card_initial_fmv))

# Show the table of distinct values
table(test$distinct_values)

Further feedback on this means that we need to figure out where diferences in final pin value come from

[Damonamajor]

OK, it's ready now.

[Damonamajor]

The changed file now includes the steps which I used to reduce the mismatched errors for pred_pin_final_fmv_round. For this, I ran stage 2 of the pipeline with the changes here and here and without. These can be understood as the modifications which would remove the Jean's proration's step from the pipeline.

Without any modifications, there are ~50,000 pins which had a different value for pred_pin_final_fmv_round. There are two key reasons why this happens. The first is that we round values before we do proration. To remedy this, I just replaced the value of pred_pin_final_fmv_round_no_prorate with pred_pin_final_fmv_round.

Then, we cap and assess land rates for prorated PINs. Since we now have modified values, these are going to be different. Because of this, I averaged the values of pred_pin_final_fmv_round_no_prorate grouped by the tieback. This is not what should be done, but does serve to match the two different runs.

At the end, there were 52 values which still did not match.

[Damonamajor]

In the meantime, I'm changing the scope of this issue to reflect the one change that we are confident we need to make to the modeling pipeline: Excluding 2-3 card PINs with different numbers of cards from the card-level prediction averaging step. This is the block code that needs to change to exclude 2-3 card PINs with different numbers of cards:

So we want the same grouping just with an additional value of pin_num_cards or the equivilent?

@jeancochrane

[Damonamajor]

@jeancochrane Can we close this until further notice?