Matrix prediction

delphi_jb1 Registered Posts: 1 ✭✭✭


I am studying a system that would make predictions for the value of a house. By using Dataiku, I already know how to go far, the software is powerful... BUT!

I realize that the software does not allow "multiple interpolated predictions (matrix prediction)"

To express myself better, here is an example of features

  • House: Price
  • House: Square meter at main floor 0
  • House: Numbers of floor
  • House: Luminosity (% of windows, regarding to the concrete walls)
  • House: Garage
  • Garden: square meter
  • Garden: secluded (closed)
  • Area: Quiet indice
  • Area: Convenience (proximity of school, shop, etc)

today, if i want to have a prediction on price or floor space, i have to create a prediction, choose the features, run it, and finally i would have my prediction... but which will be highly rigid and inflexible on the amount of basic features.
(with this prediction, tomorrow I couldn't estimate the price if I don't have all the features)

My idea:
create a new type of automated prediction, with the 9 features as input, and as output, hundreds of predictions for each feature depending on the possible configuration of the other features

  • prediction 1: price (floor space,)
  • prediction 2: price (floor space, Garage, luminosity, etc)
  • prediction 3: quiet index (price, convenience, etc)
  • ...
  • prediction 2040: convenience (all features as input)

(2040 because with 9 features, it's the number of analysis you need to cover all the configuration)

Thx in advance

1 votes

In the Backlog · Last Updated


  • Krishna
    Krishna Dataiker, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Product Ideas Manager Posts: 18 Dataiker

    Thanks for the suggestion.

  • delphi_jb
    delphi_jb Registered Posts: 3 ✭✭✭

    Here, some screens i did to have a better understanding of what i imagine:

    New analysis screen (part 1):
    You choose the new type of model (5Th)


    New analysis screen (part 2) ==> Top, Left: "All features"
    Once done, "All Features" are selected at top-left. The rest don't change


    Design ==> Metric screen:
    As the analysis will be a set of several analyzes (performed automatically), it is logical that one cannot apply a type of optimization without knowing the type of variable that one is looking for.


    Design ==> "Feature's matrix" new section:
    Here, the logic operation is explained.
    If we have an array of 4 features (A, B, C, D), then 20 predictions in total will be calculated:
    A explained by B
    A explained by B and C

    Of course, the more a feature is explained by a large number of parameters, the more accurate the prediction will be. (represented by a general diagram)


    The choice of the prediction will automatically based on the inputs:

    If I have a dataset with only "A", "D" ==> then the Matrix prediction module will automatically know which prediction to choose.

    The rest must also be thought out and developed:
    FLOW's screen: a new logo

Setup Info
      Help me…