I'm working on a project to approximate an existing model that generates prices for an asset given a large number of input variables. The model comes with ~1,000 unique example lines in batches of variations. For example, say the asset has 10 variable qualities (v1, v2... v10) that the model will use to determine three price drivers (d1, d2 and d3) which are used to compute a final price p1. The example data may have 10 lines that all have the same v1-v9 and ten different values for v10, and shows the resultant d1, d2, d3 and p1. Then there is another batch of 20 lines that hold variables v2-v10 constant and show twenty different values for v1 that produce different d1, d2, d3 and p1s for each example.

Is there any type of primer on how you would approach reverse engineering this type of predictive model? If anyone can help guide, I'd be happy to make a reasonable donation to a charity of your choosing!

Thank you!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.