Opendoor - Data Science - Sharing interview question - AMA

Jun 11, 2021 4 Comments

Just helping the community and sharing the question I received during my 1hr interview for Data Science at Opendoor.

Outcome: reject, so AMA #opendoor

----

At Opendoor, we care a lot about ensuring high valuation accuracy. We have a combination of human and algorithmic valuators who -- in both training scenarios and in production -- produce numerous valuations for different homes.

Given a manually-entered dataset of multiple valuators’ valuations (keyed by valuator_id) of different homes (keyed by home_id), evaluated against their ultimate close_prices.

We wish to understand the performance of our valuators. Define and calculate some metrics that will help us understand valuator performance. [max 10m]

How similar are our valuators’ estimates to one another? What implications does this have for reducing error via combining estimates? [max 15m]

Create and evaluate a rule that will combine valuations to achieve higher accuracy than any individual valuator. [max 15m]

Considering operational and cost constraints, design a system that will help us trade off among our key business objectives. [remaining time]

Sample data below:

home_id close_price valuator_id valuation

1 220674. V1 222988.
1 220674. V2 226617.
1 220674. V3 229378.
1 220674. V4 214982.
1 220674. V5 224865.
2 251957. V1 248071.

comments

Want to comment? LOG IN or SIGN UP
TOP 4 Comments