Ndimensional Linear regression in the RootInteractive - user interface specification https://en.wikipedia.org/wiki/Linear_regression

[miranov25]

Theoretical consideration:

• https://en.wikipedia.org/wiki/Linear_regression

User interface (scikit like)

• 1D fit
  • varX, varY, weightY
  • in root we use errY in scikit weights are use 1/erry^2
  • Variables can be used as parameters, just like for histograms.
  • Regularization as in the scikit
    • Tikhonov
    • Default as in scikit ...
• ND fit
  • [varX0, ...], varY, weightY
  • Variables can be used as parameters, just like for histograms.
  • Can be specified by multi-select
  • Regularization as in the scikit
• Nonscikit like interface. An array of 1-2D fits in the binned ND space
  • To be defined
  • Roghly should be mixed of the panda groupby and linear fit
  • Paremeters:
  • groupBy:
    • specification as in the histogram
  • [varX0, ...], varY, weightY as in the ND fit

Visualization of scikit like Value Fit

• To visualize the prediction, it is recommended to use the same granularity as the input data.
  • To achieve this, add a new alias column to the existing data source. In case of oversampling, it is advisable to use a separate data source to avoid confusion with the input data source. Careful attention should be paid to the ranges.
• Tabular visualization of the fit parameter.Columns
  • funName
  • funParIndex
  • funParValue

Visualization of the ND Value fit

[pl0xz0rz]

1D and ND fits using Tikhonov regularization already working, implemented by #325

Array of 1D fits should be simple to implement, simplest interface I can think of will be equivalent to the current interface for ND histograms