cmu statistics - nonparametric modal regression · 2015. 9. 18. · nonparametric modal regression...

49
Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department of Statistics Carnegie Mellon University August 4, 2015 Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 1 / 20

Upload: others

Post on 27-Aug-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Nonparametric Modal Regression

Yen-Chi Chen

Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman

Department of StatisticsCarnegie Mellon University

August 4, 2015

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 1 / 20

Page 2: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Motivating Examples for Modal Regression

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

Local RegressionModal Regression

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 2 / 20

Page 3: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Introduction

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 3 / 20

Page 4: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Definition for Modal Regression

We assume x ∈ K, a compact support.

Regression function–the conditional mean:

m(x) = E(Y |X = x) =

∫yp(y |x)dy .

Modal function–the conditional (local) modes:

M(x) = Mode(Y |X = x) =

{y :

d

dyp(y |x) = 0,

d2

dy2p(y |x) < 0

}.

Equivalently,

M(x) =

{y :

∂yp(x , y) = 0,

∂2

∂y2p(x , y) < 0

}.

M(x) is a multi-value function.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 4 / 20

Page 5: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Definition for Modal Regression

We assume x ∈ K, a compact support.

Regression function–the conditional mean:

m(x) = E(Y |X = x) =

∫yp(y |x)dy .

Modal function–the conditional (local) modes:

M(x) = Mode(Y |X = x) =

{y :

d

dyp(y |x) = 0,

d2

dy2p(y |x) < 0

}.

Equivalently,

M(x) =

{y :

∂yp(x , y) = 0,

∂2

∂y2p(x , y) < 0

}.

M(x) is a multi-value function.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 4 / 20

Page 6: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Definition for Modal Regression

We assume x ∈ K, a compact support.

Regression function–the conditional mean:

m(x) = E(Y |X = x) =

∫yp(y |x)dy .

Modal function–the conditional (local) modes:

M(x) = Mode(Y |X = x) =

{y :

d

dyp(y |x) = 0,

d2

dy2p(y |x) < 0

}.

Equivalently,

M(x) =

{y :

∂yp(x , y) = 0,

∂2

∂y2p(x , y) < 0

}.

M(x) is a multi-value function.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 4 / 20

Page 7: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Definition for Modal Regression

We assume x ∈ K, a compact support.

Regression function–the conditional mean:

m(x) = E(Y |X = x) =

∫yp(y |x)dy .

Modal function–the conditional (local) modes:

M(x) = Mode(Y |X = x) =

{y :

d

dyp(y |x) = 0,

d2

dy2p(y |x) < 0

}.

Equivalently,

M(x) =

{y :

∂yp(x , y) = 0,

∂2

∂y2p(x , y) < 0

}.

M(x) is a multi-value function.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 4 / 20

Page 8: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Conditional Local Modes

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 5 / 20

Page 9: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Conditional Local Modes

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 5 / 20

Page 10: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Conditional Local Modes

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 5 / 20

Page 11: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Conditional Local Modes

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 5 / 20

Page 12: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Estimator for Modal Regression

Our estimator is the plug-in from the KDE:

M̂n(x) =

{y :

∂yp̂n(x , y) = 0,

∂2

∂y2p̂n(x , y) < 0

}.

Finding conditional local modes is hard in general.

Partial mean shift: a simple algorithm for computing M̂n(x), theplug-in estimator of the KDE, from the data (Einbeck et. al. 2006).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 6 / 20

Page 13: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Estimator for Modal Regression

Our estimator is the plug-in from the KDE:

M̂n(x) =

{y :

∂yp̂n(x , y) = 0,

∂2

∂y2p̂n(x , y) < 0

}.

Finding conditional local modes is hard in general.

Partial mean shift: a simple algorithm for computing M̂n(x), theplug-in estimator of the KDE, from the data (Einbeck et. al. 2006).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 6 / 20

Page 14: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Estimator for Modal Regression

Our estimator is the plug-in from the KDE:

M̂n(x) =

{y :

∂yp̂n(x , y) = 0,

∂2

∂y2p̂n(x , y) < 0

}.

Finding conditional local modes is hard in general.

Partial mean shift: a simple algorithm for computing M̂n(x), theplug-in estimator of the KDE, from the data (Einbeck et. al. 2006).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 6 / 20

Page 15: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Example for Modal Regression

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 7 / 20

Page 16: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Example for Modal Regression

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 7 / 20

Page 17: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Asymptotic Theory

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 8 / 20

Page 18: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Error Measurement

To measure the errors, we consider the following two losses:

the pointwise loss

∆n(x) = Haus(M̂n(x),M(x)),

where Haus(A,B) is the Hausdorff distance.

the uniform loss

∆n = supx

∆n(x) = supx

Haus(M̂n(x),M(x)).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 9 / 20

Page 19: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Error Measurement

To measure the errors, we consider the following two losses:

the pointwise loss

∆n(x) = Haus(M̂n(x),M(x)),

where Haus(A,B) is the Hausdorff distance.

the uniform loss

∆n = supx

∆n(x) = supx

Haus(M̂n(x),M(x)).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 9 / 20

Page 20: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Error Measurement

To measure the errors, we consider the following two losses:

the pointwise loss

∆n(x) = Haus(M̂n(x),M(x)),

where Haus(A,B) is the Hausdorff distance.

the uniform loss

∆n = supx

∆n(x) = supx

Haus(M̂n(x),M(x)).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 9 / 20

Page 21: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Rate of Convergence

Both the pointwise and the uniform losses obey the commonnonparametric rate:

Theorem

Under regularity conditions,

∆n(x) = O(h2) + OP

(√1

nhd+3

)

∆n = O(h2) + OP

(√log n

nhd+3

).

Rate = Bias +√

Variance.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 10 / 20

Page 22: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Rate of Convergence

Both the pointwise and the uniform losses obey the commonnonparametric rate:

Theorem

Under regularity conditions,

∆n(x) = O(h2) + OP

(√1

nhd+3

)

∆n = O(h2) + OP

(√log n

nhd+3

).

Rate = Bias +√

Variance.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 10 / 20

Page 23: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Rate of Convergence

Both the pointwise and the uniform losses obey the commonnonparametric rate:

Theorem

Under regularity conditions,

∆n(x) = O(h2) + OP

(√1

nhd+3

)

∆n = O(h2) + OP

(√log n

nhd+3

).

Rate = Bias +√

Variance.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 10 / 20

Page 24: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Asymptotic Theory

To conduct statistical inferences, we ignore the bias and focus on thestochastic variation.

Theorem

Under regularity conditions,

√nhd+3∆n ≈ sup{Empirical Process} ≈ sup{Gaussian process}.√nhd+3∆n ≈ supf ∈F |B(f )| for certain function space F .

However, this is not enough for statistical inference (unknown quantities inthe Gaussian Process).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 11 / 20

Page 25: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Asymptotic Theory

To conduct statistical inferences, we ignore the bias and focus on thestochastic variation.

Theorem

Under regularity conditions,√nhd+3∆n ≈ sup{Empirical Process} ≈ sup{Gaussian process}.

√nhd+3∆n ≈ supf ∈F |B(f )| for certain function space F .

However, this is not enough for statistical inference (unknown quantities inthe Gaussian Process).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 11 / 20

Page 26: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Asymptotic Theory

To conduct statistical inferences, we ignore the bias and focus on thestochastic variation.

Theorem

Under regularity conditions,√nhd+3∆n ≈ sup{Empirical Process} ≈ sup{Gaussian process}.√nhd+3∆n ≈ supf ∈F |B(f )| for certain function space F .

However, this is not enough for statistical inference (unknown quantities inthe Gaussian Process).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 11 / 20

Page 27: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Asymptotic Theory

To conduct statistical inferences, we ignore the bias and focus on thestochastic variation.

Theorem

Under regularity conditions,√nhd+3∆n ≈ sup{Empirical Process} ≈ sup{Gaussian process}.√nhd+3∆n ≈ supf ∈F |B(f )| for certain function space F .

However, this is not enough for statistical inference (unknown quantities inthe Gaussian Process).

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 11 / 20

Page 28: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Confidence Sets

We use the bootstrap to approximate ∆n. Define another uniform metric∆̂n = supx Haus(M̂∗n(x), M̂n(x)).

Theorem

Under regularity conditions,

√nhd+3∆̂n ≈ supf ∈F |B(f )| for certain function space F .√nhd+3∆̂n ≈

√nhd+3∆n.

The set {(x , y) : y ∈ M̂n(x)⊕ t̂1−α, x ∈ K

}is an asymptotic valid confidence set for M; t̂1−α is the upper 1− αquantile of ∆̂n.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 12 / 20

Page 29: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Confidence Sets

We use the bootstrap to approximate ∆n. Define another uniform metric∆̂n = supx Haus(M̂∗n(x), M̂n(x)).

Theorem

Under regularity conditions,√nhd+3∆̂n ≈ supf ∈F |B(f )| for certain function space F .

√nhd+3∆̂n ≈

√nhd+3∆n.

The set {(x , y) : y ∈ M̂n(x)⊕ t̂1−α, x ∈ K

}is an asymptotic valid confidence set for M; t̂1−α is the upper 1− αquantile of ∆̂n.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 12 / 20

Page 30: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Confidence Sets

We use the bootstrap to approximate ∆n. Define another uniform metric∆̂n = supx Haus(M̂∗n(x), M̂n(x)).

Theorem

Under regularity conditions,√nhd+3∆̂n ≈ supf ∈F |B(f )| for certain function space F .√nhd+3∆̂n ≈

√nhd+3∆n.

The set {(x , y) : y ∈ M̂n(x)⊕ t̂1−α, x ∈ K

}is an asymptotic valid confidence set for M; t̂1−α is the upper 1− αquantile of ∆̂n.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 12 / 20

Page 31: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Confidence Sets

We use the bootstrap to approximate ∆n. Define another uniform metric∆̂n = supx Haus(M̂∗n(x), M̂n(x)).

Theorem

Under regularity conditions,√nhd+3∆̂n ≈ supf ∈F |B(f )| for certain function space F .√nhd+3∆̂n ≈

√nhd+3∆n.

The set {(x , y) : y ∈ M̂n(x)⊕ t̂1−α, x ∈ K

}is an asymptotic valid confidence set for M; t̂1−α is the upper 1− αquantile of ∆̂n.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 12 / 20

Page 32: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Example for Confidence Sets

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 13 / 20

Page 33: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Example for Confidence Sets

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 13 / 20

Page 34: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Extensions

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 14 / 20

Page 35: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Prediction Sets

We can use modal regression to construct a compact prediction set.

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

Modal Regression

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●●

● ●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

●●

●●

●●

●●

Local Regression

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 15 / 20

Page 36: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Prediction Sets

We can use modal regression to construct a compact prediction set.

●●

● ●

●●

●●

●●

Modal Regression

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

● ●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

●●

● ●

●●

● ●

●●

●●

●●

● ●

●●

●●

●●

Local Regression

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

● ●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

●●

● ●

●●

● ●

●●

●●

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 15 / 20

Page 37: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Bandwidth Selection

We can choose smoothing parameter h via minimizing the size ofprediction set.Namely, we choose

h∗ = argminh>0

Vol(P̂1−α

),

where P̂1−α is the prediction set.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 16 / 20

Page 38: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Example: Bandwidth Selection

0.00 0.05 0.10 0.15 0.20

1.0

1.5

2.0

2.5

3.0

Size of 95% Prediction interval

h

Siz

e of

pre

dict

ion

set

Optimal h

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 17 / 20

Page 39: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Regression Clustering

We can use modal regression to do ‘clustering’–exploring the hiddenstructures.

●●

●●

●●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.0

−0.

4−

0.2

0.0

0.2

0.4

0.6

0.8

X

Y

●●

● ●

●●

●●

● ●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

● ●●

●●

●●

● ●

●●

●●

●●

● ●

● ●

●●

●●

●●

●●

● ●

●●

0.0 0.2 0.4 0.6 0.8 1.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

X

Y

●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

● ●

●● ●

●●

●●

●●

● ●●●

●●

●●

●●

●●

●●●

●●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

● ●●

●●

● ●

●●

● ●

●●

● ●

●●

●●●

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 18 / 20

Page 40: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Concluding Remarks

Modal regression is very similar to mixture regression.

However, our approach is purely nonparametric–no Gaussianassumption, free from number of mixture components.

Fast to compute–no need to use EM algorithm.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 19 / 20

Page 41: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Concluding Remarks

Modal regression is very similar to mixture regression.

However, our approach is purely nonparametric–no Gaussianassumption, free from number of mixture components.

Fast to compute–no need to use EM algorithm.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 19 / 20

Page 42: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Concluding Remarks

Modal regression is very similar to mixture regression.

However, our approach is purely nonparametric–no Gaussianassumption, free from number of mixture components.

Fast to compute–no need to use EM algorithm.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 19 / 20

Page 43: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Thank you!More information and R source code can be found in

http://www.stat.cmu.edu/~yenchic

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 20 / 20

Page 44: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

reference

1. Chen, Yen-Chi, Christopher R. Genovese, and Larry Wasserman. ”Density Level Sets: Asymptotics, Inference, andVisualization.” Submitted to the Journal of American Statistical Association. arXiv preprint arXiv:1504.05438 (2015).

2. Chen, Yen-Chi, Christopher R. Genovese, and Larry Wasserman. ”Asymptotic theory for density ridges.” To appear inthe Annals of Statistics. arXiv preprint arXiv:1406.5663 (2014).

3. Chen, Yen-Chi, Christopher R. Genovese, Ryan J. Tibshirani, and Larry Wasserman. ”Nonparametric ModalRegression.” Under review of the Annals of Statistics. arXiv preprint arXiv:1412.1716 (2014).

4. Chernozhukov, Victor, Denis Chetverikov, and Kengo Kato. ”Gaussian approximation of suprema of empiricalprocesses.” The Annals of Statistics 42, no. 4 (2014): 1564-1597.

5. Chernozhukov, Victor, Denis Chetverikov, and Kengo Kato. ”Anti-concentration and honest, adaptive confidencebands.” The Annals of Statistics 42, no. 5 (2014): 1787-1818.

6. Einbeck, Jochen, and Gerhard Tutz. ”Modelling beyond regression functions: an application of multimodal regression tospeedflow data.” Journal of the Royal Statistical Society: Series C (Applied Statistics) 55, no. 4 (2006): 461-475.

7. Genovese, Christopher R., et al. ”Nonparametric ridge estimation.” The Annals of Statistics 42.4 (2014): 1511-1545.

8. Ozertem, Umut, and Deniz Erdogmus. ”Locally defined principal curves and surfaces.” The Journal of Machine

Learning Research 12 (2011): 1249-1286.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 21 / 20

Page 45: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Regularity Conditions

(A1) The joint density p ∈ BC4(Cp) for some Cp > 0.

(A2) There exists λ2 > 0 such that for any (x , y) ∈ K×K withpy (x , y) = 0, |pyy (x , y)| > λ2.

(K1) The kernel function K ∈ BC2(CK ) and satifies∫R

(K (α))2(z) dz <∞,∫Rz2K (α)(z) dz <∞,

for α = 0, 1, 2.

(K2) The collection K is a VC-type class, i.e. there exists A, v > 0 suchthat for 0 < ε < 1,

supQ

N(K, L2(Q),CK ε

)≤(A

ε

)v

,

where N(T , d , ε) is the ε-covering number for a semi-metric space(T , d) and Q is any probability measure.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 22 / 20

Page 46: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Modal Regression VS Density Ridges

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 23 / 20

Page 47: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Mixture Regression

A general mixture model:

p(y |x) =

K(x)∑j=1

πj(x)φj(y ;µj(x), σ2j (x)),

where each φj(y ;µj(x), σ2j (x)) is a density function, parametrized by a

mean µj(x) and variance σ2j (x).

Common assumptions:

(MR1) K (x) = K ,

(MR2) πj(x) = πj for each j ,

(MR3) µj(x) = βTj x for each j ,

(MR4) σ2j (x) = σ2

j for each j , and

(MR5) φj(x) is Gaussian for each j .

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 24 / 20

Page 48: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

Mixture Inference versus Modal Inference

Mixture-based Mode-basedDensity estimation Gaussian mixture Kernel density estimate

Clustering K -means Mean-shift clustering

Regression Mixture regression Modal regressionAlgorithm EM Mean-shift

Complexity parameter K (number of components) h (smoothing bandwidth)

Type Parametric model Nonparametric model

Table: Comparison for methods based on mixtures versus modes.

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 25 / 20

Page 49: CMU Statistics - Nonparametric Modal Regression · 2015. 9. 18. · Nonparametric Modal Regression Yen-Chi Chen Christopher R. Genovese Ryan J. Tibshirani Larry Wasserman Department

3D examples

X1

X2

Y

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ●●

●● ●

● ●

●●

●●●

● ●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

● ●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●

● ●

● ●

●●

●●●

●●

●●●

● ●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

● ●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●●

●●

● ●

● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ●●

●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

● ● ● ●● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

● ● ● ● ●● ● ● ●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

●● ● ● ● ●

● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

●●

●●

● ● ● ● ●● ● ● ●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

●● ● ● ● ●

● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

●●

●●

● ● ● ● ● ●● ● ● ●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●

●●

●●

●●

●●

●● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ●

●●

●●

●●

●●

●● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ●

●●

●●

●●

●●

●●

● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ● ●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ●

● ● ● ● ● ●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ●● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ●● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ●● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ●●

●● ● ● ● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ●●

●● ● ● ● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ●● ● ● ● ● ● ● ● ●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●●

●●

● ●

●●●

●●

●●

● ●

●●

● ●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

● ●

●●

● ●

●●

● ●●

●●

●●

●●

● ●

●●

●●

● ●

●●

● ●

●●

●●

● ●

●●●

●●●● ●

●●

●●

●●

● ●

●●

●●

● ●

●●

●●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●●

●●

●●

●●

● ●

X1

X2

Y

● ●●

●●● ●●● ●● ●● ●●

● ●

●●

●●●

● ●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

●●

●●

●●●

●●

●● ●● ●

●●

● ●●●

●●● ●●

● ●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ●

●●

● ●●

●●

●●

●●

●●

●●

● ●

●● ●

●●

●●

● ●

●●

●●

●●

●●

●●●

● ●●

● ●●●

●●

●● ●

● ● ●●●● ●

●●

●●●

●●●

●●

●●●● ● ●●●

●●

●●

●●● ● ●

● ●● ●●

●● ● ●●

●●● ●●● ● ●

●●●●

●● ●●●

● ●

●●

●●

●●●

●●

●●

● ●●

●● ●● ●

●●● ●

● ●●

●● ● ●

● ●

●●

●●● ●

●●

●●●

●●

●●

●●

●●●

●●

● ●●

●●

●●

● ●

●●●

●●

●●

●●

●●

●●

● ●

● ●

●●

●●

●● ●●

●●

●●

●●

●●●

●●

●●

●●●

● ●

●●

● ●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●● ●

● ●●

●●

● ●

●● ●●

●●

●●

●●

● ●●

● ● ●

● ●

●●●

●●● ●●

●●

●● ● ●

●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ●●

● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●

●● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

●●

●●

● ● ● ●● ● ● ● ●

●●

●● ● ● ● ● ● ● ● ● ●

●●

● ● ● ● ● ● ●● ● ● ● ●

●●

●●

● ● ● ● ● ● ● ●●

●●

●●

● ● ● ● ●●

●●

●●

●●

●●

● ● ● ●● ● ● ● ●

●●

●●

● ● ● ● ● ● ●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●● ● ●

● ● ● ●●

●●

●●

●●

● ● ● ● ● ● ●●

●●

●●

●●

●●

●●

●●

●● ●

● ● ● ●●

●●

● ● ● ● ● ●●

●●

●●

●●

●● ●

● ● ● ●●

●●

● ● ● ● ● ●●

●●

●●

●●

●●

●● ● ●

●●

●●

● ● ● ● ●●

●●

●●

●●

●● ● ●

●●

●●

● ● ● ● ●●

●●

●●

●● ● ● ●

●●

●●

● ● ● ● ●●

●●

●●

●● ● ● ●

●●

●●

● ● ● ● ●●

●●

●●

●● ● ● ●

●●

●●

● ● ● ●●

●●

●●

●●

●● ● ● ●

●●

●●

● ● ● ●●

●●

●●

●●

●● ● ● ●

●●

●●

● ● ● ●●

●●

●●

●●

●●

● ● ● ●●

●●

●● ● ● ●

●●

●●

●●

●●

●●

●●

● ● ● ●●

●●

● ● ● ● ●●

●●

●●

●●

●●

●●

●●

●●

● ● ● ●●

●●

●●

● ● ● ● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ● ● ● ●

●●

●●

●● ● ● ● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ● ● ●●

●●

●●

●● ● ● ● ●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

● ● ● ● ●●

●●

●●

●● ● ● ● ● ● ●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●

● ● ● ● ● ● ●●

●●

●●

●●

● ● ● ● ● ● ●●

●●

●●

●●

●● ● ● ●

●●

● ● ●●

●●

●●

● ● ● ● ● ● ● ●●

●●

●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●●

● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●●●

●●

●●

●●

● ● ● ● ●● ● ● ● ●

●●

●● ● ● ● ● ● ● ● ● ● ● ●

●● ● ● ● ● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

●● ● ● ● ●

●● ● ●

●●

●●

●●

●● ● ● ● ● ●

●●

●●

●●

●● ● ● ● ● ● ● ●

●●

●●

●●

●●

●●

● ● ● ●●

● ● ●●

●●

●●

●●

●●

● ● ●●

●●

●●

●●

●● ● ● ● ● ● ●

●●

●●

●●

●● ● ● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ● ● ● ●

●●

●●

●● ● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ● ● ●●

●●

●●

● ● ●●

●●

●●

●●

●●

●●

●●

●●

●● ● ● ● ●

●●

●●

● ● ● ●●

●●

●●

●●

●●

●●

● ● ● ●●

●●

●●

●● ● ●

●●

●●

●●

●●

●●

●● ● ●

●●

●●

● ● ●●

●●

●●

●●

●●

●● ● ●

●●

●●

● ● ●●

●●

●●

●●

● ● ● ●●

●●

● ● ●●

●●

●●

●●

● ● ● ●●

●●

● ● ●●

●●

●●

●●

● ● ● ●●

●●

● ● ●●

●●

●●

●●

● ● ● ●●

●●

● ● ●●

●●

●●

●● ● ● ● ●

●●

●●

● ● ●●

●●

●●

●●

●● ● ● ● ●

●●

●●

●●

●● ● ●

●●

●●

●●

●●

●●

● ● ● ● ●●

●●

●● ● ●

● ●●

●●

●●

●●

●●

●●

● ● ● ● ●●

●●

●● ● ● ● ● ●

●●

●●

●●

●●

● ●●

●●

●● ● ● ● ●

●●

●●

●● ● ● ● ● ● ●

●●

●●

●●

●●

●● ● ●

●●

●●

●●

●● ● ● ● ●

●●

●●

●●

●● ● ● ● ● ● ● ● ● ●

●●

●●

●● ● ● ● ● ●

●●

●●

●●

● ● ● ● ●●

●●

●●

●●

●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●

●● ● ● ● ● ● ●

●●

●●

●●●

●●

●● ●●

●●

●●●

●●

●●● ●●

●●●

●●●

●●●

●●●●

●●

●●●●●

●● ●

●●

●● ●

● ●●

● ●

●●

● ●●●

●●

●●●● ●

●●●●●●

●●●

●●●●

●●

●●●●●

● ●

●●● ●●

● ●●

●● ●●

●●

●●

●●●

●●●●

●●

●●

●●

●●●

●●●

●●●

●●● ●

● ●● ●●

●●

●●

●●●●

●●● ●●●

●●●●

● ●●

●● ●

●●

●●

● ●

●●

●●●

●● ●

● ●● ●●

●● ●

●● ●●

●●●●●

● ●

●●●

● ●

●●

●●●

●●●●

●●●●

●●●

●●●

●●

●●●●

●●

● ●●

● ●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●●●●●

●●

●●

●● ●

●●●

●●

● ●

● ●●●●

●●●●●

●● ● ●●●

●●

● ●

●● ●

●●●● ●●

●●●●

● ●●●● ●●

●●●●

●●●

●●

●●

●●●

●●

● ●●● ●●●

●●

●●

●●

● ●●

●●

●● ●

●●

●●

●●

●●

●●

● ●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

● ● ●

● ●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●●

Yen-Chi Chen (CMU-Stats) Modal Regression August 4, 2015 26 / 20