Rate a book with a program

Rate the example book with the example program:

from relativities import Program

program = Program.load("effective://example-ca-ppa")
book = program.example_book
results = program.rate(book)
results[["submission_id", "status", "reason_code", "premium"]].head(3)
#   submission_id     status reason_code premium
# 0       S000001  succeeded        None    2636
# 1       S000002  succeeded        None    5339
# 2       S000003  succeeded        None    2341

rate returns one row per submission, in book order. premium is the policy’s annual premium, as an exact Decimal. Rating runs on your machine, and your book stays there.

Count the unrated submissions

Each submission keeps its row. An unrated submission has a status other than succeeded, a reason, and NaN as its premium:

results.groupby(["status", "reason_code"], dropna=False).size()
# status          reason_code
# invalid_inputs  INPUT-TYPE       394
# refused         CSL             1056
#                 FR-FILING       1258
#                 RIDESHARE        621
# succeeded       NaN            96671

Status	Meaning
`succeeded`	The rater calculated each outcome
`refused`	The carrier declines the risk; `reason_code` is the carrier’s code
`invalid_inputs`	An input breaks `program.rater.inputs`; `reason` names the field
`rating_failed`	The rater failed on this submission

Keep the unrated submissions and their reasons in each exhibit. Results describes each column.

Summarize by your own segments

Join the results to the book by submission_id, and summarize in pandas:

rated = results.merge(book.submissions[["submission_id", "policy_id"]]).merge(book.policies)
rated.groupby("region")["premium"].agg(["count", "mean"]).astype(float).round(0)
#                   count    mean
# region
# bay_area        18722.0  3036.0
# central_coast    3908.0  2397.0
# central_valley  11700.0  2612.0
# inland_empire   11645.0  3532.0
# los_angeles     33043.0  4012.0
# north_coast      2442.0  1801.0
# sacramento       6767.0  2801.0
# san_diego        8444.0  2785.0

mean skips the NaN premiums, so each mean describes the rated policies of its region.

Rate by vehicle

Pass by="vehicles" for one row per vehicle, with the premiums the rater returns for that vehicle:

vehicles = program.rate(book, by="vehicles")
vehicles[["submission_id", "vehicle_id", "coverage_premium.collision"]].head(3)
#   submission_id vehicle_id coverage_premium.collision
# 0       S000001         V1                        562
# 1       S000001         V2                        611
# 2       S000002         V1                       1382

Explain a premium

explain lists each input the rater received, then each value it calculated, in order:

explained = program.explain(book, "S000001")
explained.query("kind == 'input'")[["name", "value"]].head(5)
#                         name     value
# 0               multi_policy      True
# 1              years_with_us         2
# 2          drivers[0].number         1
# 3  drivers[0].years_licensed        52
# 4  drivers[0].marital_status  divorced
explained.query("kind == 'calculation'")[["name", "value"]].tail(3)
#                     name value
# 201  vehicles[0].premium  1265
# 202  vehicles[1].premium  1371
# 203              premium  2636

Each row carries submission_id, so the table merges with the book and with the rows of rate. The calculation rows show each factor and premium to the total of $2,636.

Get started

Core tasks

Build and check programs

Concepts

Rater development

Python reference

Data reference

Rate a book with a program

Count the unrated submissions

Summarize by your own segments

Rate by vehicle

Explain a premium

​Count the unrated submissions

​Summarize by your own segments

​Rate by vehicle

​Explain a premium

Count the unrated submissions

Summarize by your own segments

Rate by vehicle

Explain a premium