Quantcast
Channel: Question and Answer » curves
Viewing all articles
Browse latest Browse all 18

How to model/predict discrete curve locations from numerical and categorical independent variables

$
0
0

I want to find the principal “parameters” driving the relative location of a set of $K$ discrete curves (observables, dependent variables).

The data shape is as follows:

    n  n  c  c  o
e1: a1 b1 m1 n1 x1=[x1_1,...,x1_n] 
e2: a2 b2 m2 n2 x2=[x2_1,...,x2_n] 
...
eK: aK bK mK nK xK=[xk_1,...,xk_n] 

The parameters (independent variables) are either numerical ($a_k$, $b_k$) or categorical ($m_k$, $n_k$). I would like to explain or predict the relative location of the curves with a model based on those parameters.

In a simplification of the above problem for illustration, in the case the categories are the same for all data, and only the parameter $a_k$ varies (with values $1,2,4,8, 16,32, 64$ here), the curves may look as follows:

enter image description here

In the central part of the plot ($[0.1,0.6]$), the curves lay with an order which is monotonous with $a_k$ ($x_1$ is above $x_2$, etc.). It becomes very 1D, and can be handled relatively easily. Yet, curves may cross at some locations, and other numercial and categorical variables may come into play. I welcome suggestions in the following directions, from the most generic to simplified cases.

  1. What are names of techniques and tools (if any) to address the prediction of the observed curves (on their whole range)?
  2. Are there more interesting techniques if I restrict to a range (like $[0.1,0.6]$) where curves behave nicer (are more of less above each other)?
  3. Should the curves be converted to simpler numbers (like their hierarchical rank, an area between curves)?

Viewing all articles
Browse latest Browse all 18

Trending Articles