1

I have a pretty simple .csv table with data about flat prices in several cities around the world for the years by date (city ID, city name, date, rooms, price and discount).

I want to predict flat price AND discount on the given date in the future.

The price is indicated without the discount, therefore we will assume that it hasn’t any connection with the discount and a discount CANNOT be presented in the form of (price - discount) formula. Discount can be only 5, 15 or 30%.

  1. What is the best way to set a date in table? (Can I represent date as Double in DDMMYYYY format?)
  2. What is the best way to set a discount in table? (And in which format should I get it in the output?)
  3. Can I get two outputs in one data model or should I split them into two?
  4. Which algorithms should I choose (MLRegressor, MLClassifier or something specific)?

Thank you!

instback
  • 131
  • 2
  • 15
  • 1
    Those questions aren't really specific to Create ML but are about machine learning in general. My suggestion would be to use scikit-learn to build the model and then convert it to Core ML. To learn more about machine learning, I can recommend the book *Introduction to Machine Learning with Python* by Müller and Guido from O'Reilly. – Matthijs Hollemans Feb 17 '20 at 10:58
  • Considering the limited number of data points (Dimension wise), I seriously doubt you'll ever manage to predict anything. Real estate is notoriously local, with strong subjective preferences, together with parameters that are largely independent of the tenant, ex: Room with balcony, entirely refurbished vs Room with balcony, 20 years old paint. In other words, you don't have enough columns to predict anything, IMO. Also seasonal accommodation is cyclic, so I'd choose 0...364 instead of YYYYMMDD. Algo wise, I'd look at word2vec and drop MLkIt. But that's just me. – Alex Feb 24 '20 at 06:59

0 Answers0