Scalling


Standard scalar:

Subtract the mean and scale by the sttddev

Minmax scaling:

transforms each value to a given range

Nomralization:

Log/ power transformation:

  • non linear

How to scale data:

  • Use the Training data to find the scaling parameters
    • Estimates of mean and variance
  • The , apply the transformation

When to scale data:

  • When using linear model with regulariztion
  • When using linear models without regularization
    • Model converge faster
    • Don’t need to scale if need to interpret in original units
  • When using neural networks
    • Gradient deseent
  • When using nearest-neighbor type algorithms

Regularized regression

Data : - i in indexes of n observations Linear model: - j indexes over p parameters (covariates) Loss:


TAGS