Skip to content

Variance

  • It is the difference between the error rate of training data and testing data
  • The degree of changes in the model itself with respect to changes in the training data.

  • That means the model needs to change its internal representation to incorporates the pattern in new training data.

  • high_variance

    • model is too complex, which makes it unstable and sensitive to patterns (small changes) in new data
    • High difference between training & testing data
    • Overfitting model will have #high_variance

  • low_variance

    • model is robust and simple enough and does not need many changes to incorporate new patterns in data
    • low difference between training & testing data