23
Quantile Regression Ruibin Xi

Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Quantile Regression

Ruibin Xi

Page 2: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Computation of Quantile Regression (1)

• We have

• Minimizing is equivalent to minimizing

with constraints

Page 3: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Computation of Quantile Regression (2)

• If we write

• the above problem can be written as

• Note that this is a linear programming (LP) problem

Page 4: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Linear Programming (1)

• Standard maximum problem

– Optimize a linear function subject to linear inequalities

(P) Maximize

Subject to

Page 5: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Linear Programming (2)

• Standard minimum problem

– Optimize a linear function subject to linear inequalities

(P) Minimize

Subject to

Page 6: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery Example (1)

Page 7: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery Example (2)

Page 8: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Standard LP forms

• Standard maximum (equality) form

• Standard minimum (equality) form

Page 9: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery Example: converting to standard LP form

Page 10: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Equivalent form

Page 11: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery problem: feasible region

Page 12: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery problem: Objective function

Page 13: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Brewery problem: Geometry

Page 14: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Convexity

Page 15: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Vertices (1)

Page 16: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Vertices (2)

Page 17: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(1)

Page 18: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(2)

Page 19: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(3)

Page 20: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(4)

Page 21: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(5)

Page 22: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Basic feasible solutions(6)

Page 23: Quantile Regression Lecture 1 - PKU · Brewery problem observation. Regardless of objective function coefficients, an optimal solution occurs at a vertex (0, 32) Beer (1 2, 28) (26,

Simplex algorithm