Distribution Shift

Definitions

Note

  • Distribution shift: \(p_{\text{train}}(\mathbf{x},y)\neq p_{\text{test}}(\mathbf{x},y)\)

  • Covariate shift:

    • \(p_{\text{train}}(\mathbf{x})\neq p_{\text{test}}(\mathbf{x})\)

    • \(p_{\text{train}}(y|\mathbf{x})=p_{\text{test}}(y|\mathbf{x})\)

  • Concept shift:

    • \(p_{\text{train}}(\mathbf{x})=p_{\text{test}}(\mathbf{x})\)

    • \(p_{\text{train}}(y|\mathbf{x})\neq p_{\text{test}}(y|\mathbf{x})\)

  • Label shift:

    • Only in \(y\implies\mathbf{x}\) problems.

    • \(p_{\text{train}}(y)\neq p_{\text{test}}(y)\)

    • \(p_{\text{train}}(\mathbf{x}|y)=p_{\text{test}}(\mathbf{x}|y)\)