geometric and topological data analysisfaculty.washington.edu/yenchic/talks/201709intro.pdfgeometric...

74
G T D A Yen-Chi Chen Department of Statistics University of Washington

Upload: others

Post on 30-May-2020

9 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

GEOMETRIC AND TOPOLOGICAL DATAANALYSIS

Yen-Chi Chen

Department of StatisticsUniversity of Washington

Page 2: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

1 / 28

Page 3: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 4: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 5: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 6: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 7: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 8: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric and Topological Data Analysis: Big Picture

The data can be viewed as

X1 , · · · ,Xn ∼ p ,

p is a probability density function.

Scientists are interested in geometricor topological features of p.

1 / 28

Page 9: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

The Classical Approach

◦ In all the above examples, how we estimate thegeometric/topological structures is based on plug-in estimatesfrom the kernel density estimator (KDE).

◦ Namely, we estimate the probability density function first andthen convert it into an estimator of the corresponding structure.

◦ But this idea may fail.

2 / 28

Page 10: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

The Classical Approach

◦ In all the above examples, how we estimate thegeometric/topological structures is based on plug-in estimatesfrom the kernel density estimator (KDE).

◦ Namely, we estimate the probability density function first andthen convert it into an estimator of the corresponding structure.

◦ But this idea may fail.

2 / 28

Page 11: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

The Classical Approach

◦ In all the above examples, how we estimate thegeometric/topological structures is based on plug-in estimatesfrom the kernel density estimator (KDE).

◦ Namely, we estimate the probability density function first andthen convert it into an estimator of the corresponding structure.

◦ But this idea may fail.

2 / 28

Page 12: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Failure of KDE in Analyzing Data

●●

●●●●●●●●●

●●●●●●●●

●●●●●●●●●●●

●●●●

●●

●●●

●●

●● ●

●●

●●● ● ●

●●●

●●● ●●

●●●●

●●

●●

●●●●●●

●●

●●

●●

●●

● ● ●

●●

●●●

● ●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●●

●●

●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●●●

●●

●● ●

● ●●

●●

●●●●

●●

●●

●●●●●

●●●●

●●●●

●●●●

●●●●●●

●●

●●

●●●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●● ●●

●●

●●

●●

●●●

●●●●●●●●●

●● ●●●●

●●

●●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●● ●

● ● ● ● ●●●●

● ●

●●●●●●

●●●

●●●●

●●●●

●●

●●

●●

●●

●●●

● ●●

●●

● ●● ●●

●●●●●●●●●

●●●●

●●●●

●●

●● ●●

●●●●

●●●

●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●●● ●

●●● ●

●●●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

● ●

●●

●●

●●

●●

●●●

●●

● ●

●● ●

●●

●●●●

●●

●● ●●

●●

●●

●●●●●●

●●●●

●●●●●

●●●●●●●●●

●●

●●●

●●●●●●

●●

●●

●●

●●●●●

●●●●

●●

●● ●

●●

●●

●●●●●

●●

● ●●●●●

●●●

●●

● ●

●●

● ●

● ● ● ●

●●

● ●●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●● ●

●●

●●●●●

●●●

●● ●●●●●

●●●●●●●●●●●●

●● ●

●●

●●

●●●

●●

●●

●●●

●●●● ●●

● ● ●●●

● ●

●●

●●

●●

●●● ●

●●●●

●●

●● ●

●●●

●●●●●●

●●

●●●●●●●●●●

● ●●

●●●●

● ●●

●●●●●●●●●●●●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●● ●

●●●

●●●●●●

●●

●●

●●

●●

●●

●●●●● ● ● ●●

●●

●●●●●●●●

●●●●●●●

●●● ●●●●

● ●

●●

●●

●●

●●●

●●

●●●●

●●●●

●●●●

● ●●

●●●●●

●●

●●

●●

●●

●●

●●●●●●●●

●●●●

●● ●●

●●●

● ●

●●●●●●●●

●●

●●

●●

●●●●●●●●●●

●● ●●●

●●●

●●

●●

●●

●●

● ●●

●●

●●●●●●●

●●●

●● ●

●●●

● ●●

●●

●●●●

●●●●

●●

●●

●●

●●

●●●●●●

●●

●●●

●●● ●●●●● ●●

●●

●●

●●

●●●●●●●●●●●●●●

●●

●●

●●●●● ●

●●

●●

●●●

●●

●●

●●

●●●

●●●

●●●●

● ●●●●●

●●●

●●●● ●●

● ●●

●●●●

●●

●●

● ●●●●●●●

●●●

●●●●

●●●●●●●●●

●●●●●●●●●●●

●●●

●●

●●●●●

●●

●●●

●●

●●

●●●●●

●●●● ●●●●●

●●

●●●

●●

●●●●

●●●●●●

●●●

●●●

●●

●●●●

●●

●●●●●●●

●●

●●●●●●●●●●●

● ●

●●

●●●

●● ●● ●●

●●●

●●

●●●●●●

●●●

●●●

●●●●●●●●

●●

●●

● ●●

●●●●

●●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ●●

●●●●

●●

●●

●●

● ●

● ●

●●

●●

●●

●●●●●●●●●

●●

● ●●

●●

●●

● ●

● ●

●●

●●●●●●●●

●●

●●

●●

● ● ●

●●

●●

●●

●●

● ●

●●●●●●●

●●●●

●●●

●●●●

●●●●●●

●●●●●●●●●

●●●

●●●●

●●●●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●

●●

●●●

●●●●

●●

●●

●●●

●●●

●●● ●

●●●●●●●●

● ●●●●

●●

●●

●●●

●●●●●●●●●●●

●●●

●●

●●●●

●●

●●●●

●●

●●

●● ●

● ●

●●

●●

●●

● ●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●●●

●●

●●●●●

●●

● ●

●●

●●

●●

●●

●●●●●

●●●

●●

● ●●● ●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●●

●●

●●● ●●

● ●●

●●

●●

●●

●●

●●●

●●

●●●

●●●

●●

●●●●●●●●●

●●

●●

●●●

●●

●● ●

●●

●●●

●●

●●●

●●

●●

●●●●●

●●●●●

●●

● ● ●

●●

●●

●●

●●

●●

●● ●

●●●

●●

●●

● ●●●●

●●●●●●●

●●

● ●●

●●●

●●

●●

●●

●●

●●●

●●●●

●●●

●●●●

●●

●●

●●● ●●●●●●●●

●●

●●

●●

●●

●●●●

●●

●● ●

●●●●

●●●●●●●●●●●

●●●

●●●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●●

●●

●● ● ● ●

●●

●●

●●●●

●●

●●●●●●

●●

●●

●●●

●●

●● ●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●

● ●●

● ●

●●

●●

●●

● ●

●●

●●●●

●●

●●●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●●●

●●

●●

●●

●●●

●●●●

●●●● ●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●● ● ●

●●●

●●

●●

●●●

●●

●●●

●●

●● ●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●●

●●

●●

●● ●●

●● ●●

●●

●●

●●

●●

●●

●●●

●●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●● ●

●●

●●●●●●

●●

●●

●●●●●

●●

●●●

● ●●

●●

●●●●

●● ●

● ●

● ●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●●

●●●●

● ●

● ●●

●●

●●●

●●●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●

●●

●●

● ●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●● ●

● ●● ●

●●●

●●

●●●

●●

●●

● ●●

●●

●●●

●●

● ●●●

●● ●

● ●●

●●

●●

●●

●●●

● ●

●●

●●

●●

●●●

●●

●●

●●● ●

●●●

●●

●●

●●

●●●

●●●● ●

●●

●●

●●

● ●●●

●●

●●

● ●

●●

●●●

●●●

●●

●●

●●●●

●●

0.05 0.10 0.15 0.20

−0.

45−

0.40

−0.

35−

0.30

GPS Location Records (Zoom−in)

3 / 28

Page 13: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Failure of KDE in Analyzing Data

0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

−0.

44−

0.42

−0.

40−

0.38

−0.

36−

0.34

−0.

32−

0.30 KDE

3 / 28

Page 14: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Failure of KDE in Analyzing Data

0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

−0.

44−

0.42

−0.

40−

0.38

−0.

36−

0.34

−0.

32−

0.30 Density Ranking

3 / 28

Page 15: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Failure of KDE in Analyzing Data

0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

−0.

44−

0.42

−0.

40−

0.38

−0.

36−

0.34

−0.

32−

0.30 Density Ranking + GIS data

Primary roadSecondary road

WorkplaceTownship

3 / 28

Page 16: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking: Introduction

◦ The KDE cannot detect intricate structures inside the GPS data.

◦ But the density ranking works!

◦ This comes from the fact that the underlying probability densityfunction (PDF) does not exist!

◦ Namely, our probability distribution function is a singularmeasure.

4 / 28

Page 17: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking: Introduction

◦ The KDE cannot detect intricate structures inside the GPS data.

◦ But the density ranking works!

◦ This comes from the fact that the underlying probability densityfunction (PDF) does not exist!

◦ Namely, our probability distribution function is a singularmeasure.

4 / 28

Page 18: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Definition of Density Ranking - 1

◦ Given random variables X1 , · · · ,Xn ∈ Rd , the KDE is

p(x) � 1nhd

n∑i�1

K(Xi − x

h

),

where K(·) is called the kernel function such as a Gaussian andh > 0 is called the smoothing bandwidth that controls the amountof smoothing.

◦ The KDE smoothes out the observations into small bumps andsum over all of them to obtain a PDF.

5 / 28

Page 19: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Definition of Density Ranking - 2

−0.2 0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.5

1.0

1.5

Den

sity

6 / 28

Page 20: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Definition of Density Ranking - 3

◦ The density ranking is a transformed quantity from the KDE.

◦ Instead of using the density value, we focus on the ranking of it.

◦ The formal definition of density ranking is

α(x) � 1n

n∑i�1

I�p(x) ≥ p(Xi)�

� ratio of observations’ density below the density of point x.

◦ Namely, α(x) � 0.3 implies that the (estimated) density of point xis above the (estimated) density of 30% of all observations.

7 / 28

Page 21: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Definition of Density Ranking - 3

◦ The density ranking is a transformed quantity from the KDE.

◦ Instead of using the density value, we focus on the ranking of it.

◦ The formal definition of density ranking is

α(x) � 1n

n∑i�1

I�p(x) ≥ p(Xi)�

� ratio of observations’ density below the density of point x.

◦ Namely, α(x) � 0.3 implies that the (estimated) density of point xis above the (estimated) density of 30% of all observations.

7 / 28

Page 22: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Definition of Density Ranking - 3

◦ The density ranking is a transformed quantity from the KDE.

◦ Instead of using the density value, we focus on the ranking of it.

◦ The formal definition of density ranking is

α(x) � 1n

n∑i�1

I�p(x) ≥ p(Xi)�

� ratio of observations’ density below the density of point x.

◦ Namely, α(x) � 0.3 implies that the (estimated) density of point xis above the (estimated) density of 30% of all observations.

7 / 28

Page 23: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Property of Density Ranking

◦ For an observation Xmax with α(Xmax) � 1, then it means

p(Xmax) � max�p(X1), · · · , p(Xn) .

◦ Similarly, for an observation Xmin with α(Xmin) � 0,

p(Xmin) � min�p(X1), · · · , p(Xn) .

◦ If an observation X` satisfies α(X`) � 0.25, this means that theranking of density at X` is the 25%.

◦ Moreover, for any pairs of points x1 , x2,

p(x1) > p(x2) �⇒ α(x1) > α(x2)p(x1) < p(x2) �⇒ α(x1) < α(x2)p(x1) � p(x2) �⇒ α(x1) � α(x2)

8 / 28

Page 24: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Property of Density Ranking

◦ For an observation Xmax with α(Xmax) � 1, then it means

p(Xmax) � max�p(X1), · · · , p(Xn) .

◦ Similarly, for an observation Xmin with α(Xmin) � 0,

p(Xmin) � min�p(X1), · · · , p(Xn) .

◦ If an observation X` satisfies α(X`) � 0.25, this means that theranking of density at X` is the 25%.

◦ Moreover, for any pairs of points x1 , x2,

p(x1) > p(x2) �⇒ α(x1) > α(x2)p(x1) < p(x2) �⇒ α(x1) < α(x2)p(x1) � p(x2) �⇒ α(x1) � α(x2)

8 / 28

Page 25: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Property of Density Ranking

◦ For an observation Xmax with α(Xmax) � 1, then it means

p(Xmax) � max�p(X1), · · · , p(Xn) .

◦ Similarly, for an observation Xmin with α(Xmin) � 0,

p(Xmin) � min�p(X1), · · · , p(Xn) .

◦ If an observation X` satisfies α(X`) � 0.25, this means that theranking of density at X` is the 25%.

◦ Moreover, for any pairs of points x1 , x2,

p(x1) > p(x2) �⇒ α(x1) > α(x2)p(x1) < p(x2) �⇒ α(x1) < α(x2)p(x1) � p(x2) �⇒ α(x1) � α(x2)

8 / 28

Page 26: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Property of Density Ranking

◦ For an observation Xmax with α(Xmax) � 1, then it means

p(Xmax) � max�p(X1), · · · , p(Xn) .

◦ Similarly, for an observation Xmin with α(Xmin) � 0,

p(Xmin) � min�p(X1), · · · , p(Xn) .

◦ If an observation X` satisfies α(X`) � 0.25, this means that theranking of density at X` is the 25%.

◦ Moreover, for any pairs of points x1 , x2,

p(x1) > p(x2) �⇒ α(x1) > α(x2)p(x1) < p(x2) �⇒ α(x1) < α(x2)p(x1) � p(x2) �⇒ α(x1) � α(x2)

8 / 28

Page 27: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking as an Estimator

◦ Density ranking α(x) can be viewed as an estimator to a functionof the underlying population distribution.

◦ When the distribution function has a PDF, the population versionof density ranking is defined as follows.

◦ Assume X1 , · · · ,Xn is a random sample from an unknowndistribution function P with a PDF p.

◦ Then the population version of α(x) isα(x) � P(p(x) ≥ p(X1)).

◦ Under regularity conditions,∫ �α(x) − α(x)�2 dP(x) P

→ 0, supx

�α(x) − α(x)� P

→ 0.

9 / 28

Page 28: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking as an Estimator

◦ Density ranking α(x) can be viewed as an estimator to a functionof the underlying population distribution.

◦ When the distribution function has a PDF, the population versionof density ranking is defined as follows.

◦ Assume X1 , · · · ,Xn is a random sample from an unknowndistribution function P with a PDF p.

◦ Then the population version of α(x) isα(x) � P(p(x) ≥ p(X1)).

◦ Under regularity conditions,∫ �α(x) − α(x)�2 dP(x) P

→ 0, supx

�α(x) − α(x)� P

→ 0.

9 / 28

Page 29: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking as an Estimator

◦ Density ranking α(x) can be viewed as an estimator to a functionof the underlying population distribution.

◦ When the distribution function has a PDF, the population versionof density ranking is defined as follows.

◦ Assume X1 , · · · ,Xn is a random sample from an unknowndistribution function P with a PDF p.

◦ Then the population version of α(x) isα(x) � P(p(x) ≥ p(X1)).

◦ Under regularity conditions,∫ �α(x) − α(x)�2 dP(x) P

→ 0, supx

�α(x) − α(x)� P

→ 0.

9 / 28

Page 30: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking as an Estimator

◦ Density ranking α(x) can be viewed as an estimator to a functionof the underlying population distribution.

◦ When the distribution function has a PDF, the population versionof density ranking is defined as follows.

◦ Assume X1 , · · · ,Xn is a random sample from an unknowndistribution function P with a PDF p.

◦ Then the population version of α(x) isα(x) � P(p(x) ≥ p(X1)).

◦ Under regularity conditions,∫ �α(x) − α(x)�2 dP(x) P

→ 0, supx

�α(x) − α(x)� P

→ 0.

9 / 28

Page 31: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking in Singular Measure

◦ Why density ranking works in GPS data but KDE fails is probablydue to the fact that density ranking is a consistent estimator evenwhen the density does not exist!

◦ To generalize population density ranking to a singular measure,we introduce the concept of geometric density.

◦ Let Cd be the volume of a d dimensional ball andB(x , r) � {y : ‖x − y‖ ≤ r}.

◦ For any integer s, we define

Hs(x) � limr→0

P(B(x , r))Cs rs .

◦ For a point x, we then define

τ(x) � max{s ≤ d : Hs(x) < ∞}, ρ(x) � Hτ(x)(x).

10 / 28

Page 32: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking in Singular Measure

◦ Why density ranking works in GPS data but KDE fails is probablydue to the fact that density ranking is a consistent estimator evenwhen the density does not exist!

◦ To generalize population density ranking to a singular measure,we introduce the concept of geometric density.

◦ Let Cd be the volume of a d dimensional ball andB(x , r) � {y : ‖x − y‖ ≤ r}.

◦ For any integer s, we define

Hs(x) � limr→0

P(B(x , r))Cs rs .

◦ For a point x, we then define

τ(x) � max{s ≤ d : Hs(x) < ∞}, ρ(x) � Hτ(x)(x).

10 / 28

Page 33: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking in Singular Measure

◦ Why density ranking works in GPS data but KDE fails is probablydue to the fact that density ranking is a consistent estimator evenwhen the density does not exist!

◦ To generalize population density ranking to a singular measure,we introduce the concept of geometric density.

◦ Let Cd be the volume of a d dimensional ball andB(x , r) � {y : ‖x − y‖ ≤ r}.

◦ For any integer s, we define

Hs(x) � limr→0

P(B(x , r))Cs rs .

◦ For a point x, we then define

τ(x) � max{s ≤ d : Hs(x) < ∞}, ρ(x) � Hτ(x)(x).

10 / 28

Page 34: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking in Singular Measure

◦ Why density ranking works in GPS data but KDE fails is probablydue to the fact that density ranking is a consistent estimator evenwhen the density does not exist!

◦ To generalize population density ranking to a singular measure,we introduce the concept of geometric density.

◦ Let Cd be the volume of a d dimensional ball andB(x , r) � {y : ‖x − y‖ ≤ r}.

◦ For any integer s, we define

Hs(x) � limr→0

P(B(x , r))Cs rs .

◦ For a point x, we then define

τ(x) � max{s ≤ d : Hs(x) < ∞}, ρ(x) � Hτ(x)(x).

10 / 28

Page 35: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking in Singular Measure

◦ Why density ranking works in GPS data but KDE fails is probablydue to the fact that density ranking is a consistent estimator evenwhen the density does not exist!

◦ To generalize population density ranking to a singular measure,we introduce the concept of geometric density.

◦ Let Cd be the volume of a d dimensional ball andB(x , r) � {y : ‖x − y‖ ≤ r}.

◦ For any integer s, we define

Hs(x) � limr→0

P(B(x , r))Cs rs .

◦ For a point x, we then define

τ(x) � max{s ≤ d : Hs(x) < ∞}, ρ(x) � Hτ(x)(x).10 / 28

Page 36: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric Density: Example - 1

◦ Assume the distribution function P is a mixture of a 2D uniformdistribution within [−1, 1]2, a 1D uniform distribution over thering {(x , y) : x2 + y2 � 0.52}, and a point mass at (0.5, 0), then thesupport can be partitioned as follows:

11 / 28

Page 37: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric Density: Example - 2

◦ Orange region: τ(x) � 2 .◦ Red region: τ(x) � 1 .◦ Blue region: τ(x) � 0 .

12 / 28

Page 38: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric Density and Ranking

◦ The function τ(x)measures the dimension of P at point x.

◦ We can then use τ and ρ to compare any pairs of points andconstruct a ranking.

◦ For two points x1 , x2, we define an ordering such that x1 �τ,ρ x2 if

τ(x1) < τ(x2), or τ(x1) � τ(x2), ρ(x1) > ρ(x2).◦ Namely, we first compare the dimension of the two points, the

lower dimensional structure wins. If they are on regions of thesame dimension, we then compare the density of that dimension.

13 / 28

Page 39: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric Density and Ranking

◦ The function τ(x)measures the dimension of P at point x.

◦ We can then use τ and ρ to compare any pairs of points andconstruct a ranking.

◦ For two points x1 , x2, we define an ordering such that x1 �τ,ρ x2 if

τ(x1) < τ(x2), or τ(x1) � τ(x2), ρ(x1) > ρ(x2).

◦ Namely, we first compare the dimension of the two points, thelower dimensional structure wins. If they are on regions of thesame dimension, we then compare the density of that dimension.

13 / 28

Page 40: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Geometric Density and Ranking

◦ The function τ(x)measures the dimension of P at point x.

◦ We can then use τ and ρ to compare any pairs of points andconstruct a ranking.

◦ For two points x1 , x2, we define an ordering such that x1 �τ,ρ x2 if

τ(x1) < τ(x2), or τ(x1) � τ(x2), ρ(x1) > ρ(x2).◦ Namely, we first compare the dimension of the two points, the

lower dimensional structure wins. If they are on regions of thesame dimension, we then compare the density of that dimension.

13 / 28

Page 41: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Constructing Density Ranking using Geometric Density

◦ Using the ordering �τ,ρ, we then define the population densityranking as

α(x) � P(x �τ,ρ X1)

◦ When the PDF exists, the ordering �τ,ρ equals to �d ,p so

α(x) � P(x �d ,p X1) � P(p(x) ≥ p(X1)),which recovers our original definition.

14 / 28

Page 42: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Constructing Density Ranking using Geometric Density

◦ Using the ordering �τ,ρ, we then define the population densityranking as

α(x) � P(x �τ,ρ X1)◦ When the PDF exists, the ordering �τ,ρ equals to �d ,p so

α(x) � P(x �d ,p X1) � P(p(x) ≥ p(X1)),which recovers our original definition.

14 / 28

Page 43: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Convergence under Singular Measure

◦ When P is a singular distribution and satisfies certain regularityconditions, ∫ �

α(x) − α(x)�2 dP(x) P→ 0

but no guarantee for the convergence of supx�α(x) − α(x)�.

◦ Example of non-convergence of supreme norm: points very closeto a lower dimensional structure will not converge.

15 / 28

Page 44: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Convergence under Singular Measure

◦ When P is a singular distribution and satisfies certain regularityconditions, ∫ �

α(x) − α(x)�2 dP(x) P→ 0

but no guarantee for the convergence of supx�α(x) − α(x)�.

◦ Example of non-convergence of supreme norm: points very closeto a lower dimensional structure will not converge.

15 / 28

Page 45: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking and Cluster Tree - 1

◦ Cluster tree is a technique to summarize a function using a tree.◦ When the PDF exists, the cluster tree of a PDF and the cluster tree

of the corresponding density ranking has the same tree topology.

p(x)

x

p(x)

x

◦ The idea of building a cluster tree of a function f relies onmatching the connecting components of level sets {x : f (x) ≥ λ}when we vary the level λ.

16 / 28

Page 46: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking and Cluster Tree - 2

◦ Using the level sets of α(x) or α(x), we can define the cluster treeof the density ranking and the population density ranking.

◦ When the distribution function is singular and satisfies certainregularity conditions, the cluster tree of α(x) converges to thecluster tree of α(x).

17 / 28

Page 47: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking and Cluster Tree: Example

Here the population distribution function is a mixture of a 1Dstandard normal distribution and a point mass at 2. We consider threesample sizes: n � 5 × 103 , 5 × 105 , 5 × 107.

−3 −2 −1 0 1 2 3

0.0

0.1

0.2

0.3

0.4

0.5

0.6

N = 5000 ; Bandwidth = 0.21

Density

−3 −2 −1 0 1 2 3

0.0

0.5

1.0

1.5 N = 5e+05 ; Bandwidth = 0.08

Density

−3 −2 −1 0 1 2 3

0.0

0.5

1.0

1.5

2.0

2.5

3.0

N = 5e+07 ; Bandwidth = 0.03

Density

0.0

0.2

0.4

0.6

0.8

1.0

α(x)

0.0

0.2

0.4

0.6

0.8

1.0

α(x)

0.0

0.2

0.4

0.6

0.8

1.0

α(x)

18 / 28

Page 48: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Application of Density Ranking: GPS dataset - 1

●●

●●●

●●

●●

● ●

●●

●●●●

●●

●●

●● ●

●●●● ●

●●●

●●

●●

●●

●●●

●●● ● ●●

●● ●

● ●

●●

●●

●●●

●●●

●● ● ●

●●

●●●●●

●●

●●

●●

●●●

●●●

●●

●●●●● ●

●●

●● ●●

●● ●● ● ●●● ●●●●● ●

●●

●● ●●

●●

●●●●

●●●

●●●

●●

●●●●●

●●●●

●●

●● ● ●

●●

●●●

●●

●●

●● ●●●

●●

● ● ●

●● ●

●● ●

●●

●●●

●● ●

●●●

●●●

●●●

●●

● ●

●●●●

●●

●●

●● ●

●●●●

●●

●● ●

●●

●●●●●●●●

●●

●●●●● ● ●

●●

● ●●

●●●

●●

●●

●●●

●●●●●

●●●●●●●●

●●●●

● ●

● ●

●●

●●

● ● ● ● ● ●●●●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●●●●●

●●

●●

●●●

●●●●

●●●●

●●●

●●●●●

●●

●●●● ● ● ●

●●

●●

●● ●●●

●●

●●

●●

●●● ●

●●

●●

●●●●

●●●

● ●

● ●

●●

●●

● ●●

●●

●●●

●●●●

●●●

●●

●●●

●●

●●

●●

●●●

● ●

●●

●●●●

●●

●●

●●●●●

●●

●●

●●

● ●

●●●

●● ●●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●

●● ● ●

●●●

●●

● ●

●●●●●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●

●● ● ● ● ●●

●●

●●●

●●

●●

●●

●●●●●

●●●●

●●

●●

●●●●

●●

●●

●●●

●●

●●●

●●●●

●●

●●●●●

●●●

●●

●●

●●● ● ●●

●●●

● ●

●●

●●

● ●●

●●

●● ●

●●

● ●●●

●●●

● ●●

●●● ●● ●

●●

●●

●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

● ●●

●●

●●●

●● ●

●●●●

●●● ●●

●●●●●●●●

●●●

●●

● ●●

●●● ●●●

●● ●

●●

● ● ●●

●●●

●●

● ●● ●●● ●●● ●

●●

●●

●●●●●●●

●●

●●●●●●●● ●●●

●●

●●

●●●●● ●

●●

●●● ●

● ●

●● ●●● ●●●

●●

●●

●●●

●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●●●

●● ●

●●●●●●●

●●●●●●●

●●

●●

●●●●● ●

●●●

●●●●● ●

●●

●●●

●●● ●

●●●●

●●

●●

●●● ●

●●●

●●●

●●

●●

●●●

●●

●●

●●●●

●●●●●

●●

●●●●●

● ● ●●

● ●

●●

●●

●●

●●

● ●

●●●

●●

●●●●●

●● ●● ● ●● ●●

●●●

●●●●●●●●●

●●

● ● ●●

●●

●●

●●●●●

●●

●●●●

●●

●●

●● ● ●●●●

●● ● ● ● ●

● ●

●●

●●

●●

●●

●●●

●●

●●●●

●● ●●●

●●

●●●●●

●●●●

●● ●

●●

●●●

●●

●●

●● ●●

●●

●●

●●● ●● ●

●●

●●● ●●

●●●

●●

●●●●●●

●●

●●

●●●

● ●●●

●●●

●●

●●●

●●●

●●

●●●

●● ●● ● ●●● ●

●●

●●●●●●

●●

●●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●● ●

●●

●●●

●● ●

●●

●●●●

●●

● ●●

●●● ● ●●

●●

●●●

●●

●●

●●●

●●

●●

●●●●●

●●

●●

● ●●●

●●●

● ●

●●

●●

●●●●

●●●●

●●●

●● ●

●●

● ●

●●

●● ●

●●●

●●●●●

●● ●●

●●

●●● ●

● ● ●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●●

● ●

●●

●●●

● ●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●● ●

●●

●●

● ●

●●

●●●

● ●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●●

●●

●● ●

●●●

● ●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●● ●

●●

●●●●●

●●●

●●

●●●●

●● ●

●●

●●●

●●

●●

●●●●

●●

●●

●● ●● ●

●●

●●

●●

●●

●●●

● ●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●●

●●●

●●

●● ●●● ●

●●

●●

●●●●

●● ●

●●

●●

●●

●●●●

●●

●●

●●●

●●●● ●

●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

● ●

●●●● ●● ●● ●● ●●

●●

●●

●●

●●

●●

●●

●● ●

●● ●

●●

●●●

●●

●●●

●●

●●

●●

●●●

● ●●

●●

●●●●

● ●

●●

●●

●●●●●

● ●

●●

● ●

●● ●●

●●

●●● ●

●●

●●● ●

●●

●●

●● ●

● ●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●

●● ● ●

●●

●●

●●

●●

●●●

● ●

● ●●

●●

●●

●●

●●

●●

●● ●● ●● ●● ●

●●●

●●●●

●● ●●

●●

●●●●

●●

●●

● ●

●●●

●●

●●●

●●

●●

●●

●●●●

● ●●

●●●

●● ●●

●●●

●●●

●●●

●●

●●

●●●

●●●

●●

●●

●●●

● ●●●

●●

●●

●●

●●

●●

●●

●●

● ●●●

●●

●●

●●

● ●●

●●● ●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

● ●

●●

●●●

●●

●●●●

●●

●●●●

●● ●

●●

●●

●●●

●●

●● ● ●

●●

● ●

●●

●● ●

●●

●●

●●

●● ●●

●●

●●

●●●●

●●

●●

●●

●●

●●●

●●●●

●●

●●● ● ●●●

●●

● ●●

●●

●●●

●●

●●

●●●

● ●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●● ●

● ●● ●

● ●●

● ● ●

●●

●●●●●●

●●

●●

●●

●●

●●●

●●

●●●●

●●●●●●●●

●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●

●●

●●

●●

●●●●●●●●●●●●●

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

●●●●●●

●●●

●● ●

●●●●●●●●●●●●●●●●

● ●●

●●●

●●●

●●

●●

●●●●

●●

●●●

●●

●● ●

●●●

●●

●●●

●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●

● ●●

●●●●

●●

●●●

●●●●●●●●

●●

● ●●

●●●●●

● ●

● ●●

●●●

●●●●●●

●●●

●●●●●●●●

●●●

●●●●●

●●

●●

●●

●●●

●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●

●●

●●

●●●●●●●●●●●●●●

●●● ●

●●●●●●● ●●●●●●●

●●●●●●

●●●●●●●●

●●●●●●●●●●

●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●

●●

●●●● ●

●●●●●●●

●●

● ●●

●●●●

●●●●●

● ●●

●●●●●●●●

●●

●●●●●●●●●●●●●

●●

●●●

●●●●●●●●●●

●●

●●

●●●●●●●●

●●●●●●●●●

●●

●●●●●●●●●●●●

● ●

●●●●●●●●●●●●●●●●●●●

●●●●

●●

● ●

●●

●●●●

● ●

●●● ●●● ●

●●●●●●●●

●●

●●●●●●

●●

●●●●●●●●

●●●

●●●●●●●●●●●●●●●●●●●

●●●●●●

●●●●

●●

●●●●●●●

●●

●●●●●●●●●

●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

● ●●

● ●

●●

● ●

Individual: 1

●●●●

● ●●●

●●

●●

●● ●●●

● ● ●●●

●●

●● ●

●●●●

● ●

●●

●●

●●●● ●●

●●●● ●

●●●●●●●● ●●●● ● ●●

●●●

●● ●●● ●●

●● ●

●●

●●

●●●●

●● ●●●●

●●●

●●●●

●●●

●● ●

●● ●●●●

●●●●

●●

●●

●●● ● ●

●● ●●

●●●●●

●●●

●●●●

● ●

●●●●●●●

●●

●●●

●● ● ●●

●●

●●●

●●

●●

● ●●●●●●●

●●

●●●● ●

●●

●●

●●

●●●

●●

●●

●●

●●

● ●●

●●●●

●●●

●●●

●●● ●

●●●

●●●

●●

●●●

●●

●●●

●● ●

●●

●●

●●

●●

●●●

●● ●●●●●

●●

●●

●●●

●●●●

●●

●●

●●●●●●●

●●●

●● ●●

●●

●●

●●●●● ●● ●●

●●

●●

●●●●● ●

● ●

●●

●●

●●

●●

● ●

●●●●● ●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●● ●●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

● ●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●●●

●●●●

●●●

●●

●●●

●●

●●

●●

●●

● ●●●●●

●●●●●

●●

●●●●

●● ● ● ●

●●●●

●●

●●

●●●●

●●

●●

●●

● ●●● ● ●

●●

●●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●●●●●●●●●●●●●●

●●●

●●●●

●●

●● ●●●

●●●●●

●●●

●●●●

●●

●● ●●●

● ●

● ●

● ●●●

●●

●●

●●

●●

●●

●●

●●●●● ●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●●

●●●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●●●

●●

● ●

●●

●●●

●●

●●

● ●●

●●

●●●

●●●●●

●●

●●

●●

●●

●●

●●●●● ●

● ●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

● ●

●●

●●

●●

● ● ●●

●●

●●

●●

● ●

●● ●

● ●● ● ●

●●

●●●

●●

●●

● ●

● ●

●●

●●

●● ●

● ●

●●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

● ●●●

●●● ●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

● ●●

● ●

● ●

●●

●●

●●

●●●

● ●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ● ●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●● ●

●●●

● ●● ●●

●●

●●

●●●

●●

●●

●● ●●●

●●

●●●

●●

●●●

●●●

●●●

●●

●●

● ●

●●

●●

●●

●●●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●●●

●●●

● ●

●●●

●●

●●

●●●

●●

● ●

●●

●●

●● ●

●●●●

●●

●●●

● ●

●●

● ●

●●●●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●● ●

●●

●●

●●●

● ● ●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●●

●●

●●●●

●●

●●

●●●●

●●

●●

●●

● ●●●●●●

●●

●●●

●●

●●

●●●●● ● ●

● ●

●●

●● ●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●● ●●

●●

●●

●●

●● ●●

●●

●●

●●

●●

● ●

●●●

●● ●

●●

●●

●●●●●●●●

●●

●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●

●●●

● ●●

●●

●●

●●

● ●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●

●●●

●●●●

●●

● ●

●●●

●●●

●●

●●●

●●

●●

●●

●●●●●

● ●

●●

●●●●●

●●

●●

●●

●●●●

●●●

●●● ●

●●

●●●●●●

●●●●●

●●

●●●●●●●●●●●●●●●●●●●●●

●●

●●

●●

●●●●

●● ●

●●

●●●●

●●

●●

●●

●●

●●

●●●●●

●●●

●●●●

● ●

●●●●

●●●●●

●●

●●

●●

●●●●

●●

●●●●

●●●●●●●

●●●●●●●●●

●●●

●●●●●●●● ●●●

●●●●●● ●●●●●●● ●●●●

●●

●●●●●●●●●

●●●

●●●●●●●

●● ●●●●● ●●●●●●●●●●●

●●

●●

●●●

●●

●●●

●●

●●●●

●●●

● ●●●

●●●●●●●●●●●

●●

●●●

●●

●●●●●●

●●

●●

●●●● ●●●●●●● ●●●●

●●●

●●

●●●

●●●●●

●●●

●●

●●

●●●●●

●●●

●●●●

●●

●●

●●

●●

●●●

●●●

●●●

● ●●●

●●●

●●●●●●●

●●●●

●● ●

●●

●●

Individual: 2

●●

●●

●●●

●●

●●●●

● ●

●●

●●

●●● ●● ●●

●●

●●●●●

● ●●

●●●

●●

●●

●●●

●●

●●●●

●●

●●●

● ●

●●

● ● ● ●

●● ●●●●

● ●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●●●●●●

●●

●●

●●●●●●

●●

●●

●●●

●●

● ●●

●●

●●●●

●●

●●

●●●●

●●●

●●●

●●

●●

●●●

●●

●● ●●

●●

●●●●

●●

●●●●

●●

●●

●●● ●●●●

●●●

●● ●●

●●

● ● ●●

●●●●

●●

●●

●●●●

●●

●●

●●

●● ●

● ●

●●●●

●●

●●●

● ● ●●

●●

●●●

●●

●●●

●●

●●

●●

●●●

●●●

●●

●●

●●●●

● ●

●●●

●●

●●●● ●●

●●

●●

●●

●●●

●●●

●●●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

● ●

●●●●

●●●

●●●

● ●

●●

●●

●●●●●●●●●

●●●●

●●

●●

●●

●●

●●

●●●● ●●

●●●

●●●●●

●●●●

●●

●●

●●●●

●● ●

●●●●●

●●

●●●●●●●

●●●

●●●

●●

● ●●

●●

●●

● ●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●● ●

● ●●●

● ●●

●●

●●●

●● ●●●●

●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●●

●●

●●

●●●●

● ●●

●●

● ● ●

●●●

●●●

●●

●●●●

●●●●

●●

●●●●●

●●●●●

●● ●

●●

●●

●●

●●●●●●●●

●●●

●●

●●

●●

●●

●●●●●●●●●●●●

●●●●●●●

●●

●●

●●●●●

●●

●●●●

● ●●

●●

●●

● ●● ● ●●

● ●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●●●●●●

● ●●●

●●

●●

●●

●●●

●●

●●●

● ●

● ●

●●●

●●

●●

●●

●●

● ●●●

●●

●●●

●●●

●●

●●

●●

●●

●●●●●

●●

●●●●●●

●●●

●● ●●

●●●

●●●●●

●●●

●●

●●

●●●●

●●

●●●

●●●

●●

●●●●

●●●●

●●

●●

●●

● ●

●●

● ●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●● ●●

●●

●●●

● ● ●●

●●

●●

●●●●●●●

●●● ●●

●●

●●

●●

●●

● ●●

●●

●●●●

●●

●●●

●●● ●●● ●●●

●●

●●●

● ●●

●●

●●●

●●

●●

●● ●●

●●●

●●

●●●●●●

●●

● ●●●●

●●● ●● ●

●●

● ●

●●

●●

●●●● ●

●●●●

●●●

● ●●

●●●

●●

●●

●●●

●●●

●●● ●●

● ● ● ●●●

●●●●●●

●●●

●●●

●●

●●●

●●●●●

●●●

●●●

●● ●

●●●●

●●●●●●●● ●● ●

●●●

●●●●

●●●●●

●●

●●

●●●

●●●●●●● ●

●●●

●●

●●●●●

●●●

●●

●●●

●●

●●

● ●

●●●●●●

●●

●●

●●

●●

●●

●●

●●

● ●●

● ●●

●●

●● ●●

●●

●●

●●●●●●

●●●

●●

●●●●

●●

●●

●●●●●

●●

●●●

●●

●●●

●●

●●

●●

●●●

●●● ●●●

●●

●●

●●● ●●●

●●●

●●●

●●

●●●

●● ●●●

●●

●●

● ●●

●●

●●

●● ●●

●●●

●●

●●●

●●

●●

● ●

●●

● ●

●●●●●

●●●●●

●●

●●

●●●●●●●

●●●

●●●●

●●

●●

● ●

●●●

●●

●●●●●●●●

●●●●●●

●●●

●●

●●●

●●●●●●

●●●●●●

●●

●●●●

●●●●●●●●●●

●●●●

●●

●● ●● ●

●●●

●●

●●●

●●

●●

● ●●

●●

●●

●●

●●

●●●●

●●

●●

● ●●●●●

● ●

●●

●●●

●●●

●●

● ●

●●●

●●

●●

● ●

●●●

●●●

●●

●●

●●●●●●●

●●

●●●

●●

● ●●

●●●

●●●● ●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●

●●

●●

●●

●● ● ●

●● ● ●

●●

●●●

●●●

●●●

●●

●●

●●●

●●

● ●

● ●●● ●

● ●

●●

●●●

●●

●●

● ●

● ●●●●

●●●

●●●●

●●●

●●

●●●

●● ●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●● ●

●●

●●

●●

● ●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●●

●●●

●●

●●

● ●

●●

●●

●●

●● ●

●●

●●

●●

● ●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●● ●●

●●●

●●

● ●

● ●

● ●

●●

●●

● ●●

●●●

●●●●

●●

●●

● ●● ●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●●

●●

●●

●●

●●

● ●●

●●

●●

●● ●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

● ●●

● ●●

●●

●● ●●●●

●●●

●●

●●●

●●

● ●●

●●

●●

● ●

●●●

●●

●●●

● ●

● ●●

●●

●●

●●

●●

●●●

●●

● ●●●

●●

●●●

●●

●●●●

●●●

●●

●●

●●

●●●

● ●●● ●

●●

●●

●●

●●

●●

●●

● ●●

●●●

● ●

●●●

●●

●●

●●●

●●

●●

●●

● ●

●●

●●● ●

●●

●●

●●

●●●

●●

●● ● ●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●●

●●●

● ●●●

●●

● ●●

●●

●●●

● ●●

●●

●●● ●

●●

●● ●

●●

●●●●

● ●

●●

● ●

●●

●●

● ●

●● ●

●●

●● ●

●●

●●

●● ●

●●

●●

●●●●● ●

●●●

●●

● ●

● ●●●

●●

●●

●● ●

●●

●●

●●● ●●

●●

●●● ●

●●●

●●

●●

●●

●●

●●

●●●● ●

●●

●●●

●●

●●

● ●

●●

●●

●● ●

●●●● ●●

●●

●●●●

●●

●●

● ●●

● ●

●●

●●

●●

●●● ●

●●●

●●●

●●●●●

●●

●●

●●

● ● ●●●

●●

●●●●

●●●

●●●

●●

●● ●

●●

●●

●●

●●

●●● ●●●

●●

●●●

●● ●

● ●

● ●●●

●●●

●●

●●●●

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

●●

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

●●

●●●●●●

●●●●●

●●

●●●●●●●●

●●

●●●●

●●●

●●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●●●

●●● ●

●●

●●

●●

●●

● ●

●●

●●

●●●

●●

●●●●●●●●

●●●

●●

●●

●●

●●

●●●●●

●●

●●●

●●

●● ●

●●●●

●●

●●

●●●●●

●●

●●

●●

●●●●●

●●

●●

●●

●● ● ●

●●

●●● ●●

●●●

●●●

●●

●●

●●

●●

● ●

●●●

● ● ●●

●●●

●●●●●●●●●●●●●●●●

●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●

●● ●●

●●

●●●

●●●●●●●●●●●●●●●●●●●●●●●●●

●●●

●● ●

●●●●

●●

●●●●●●●●●●●●●●

●●

●●●

●●

●●

●●

●●●●●

●● ●●●●● ●

●●

●●●●

●●

●●

●●

●● ●●●

● ● ●● ● ●●●●●●●●●●

●●● ●●

●●

●●

●●

●●●●●●

●●

●●●

●●

●●

●●●●●●●●

●●●

●●

●●●●●●●●●●●●●

●●

●●●

●●●●●●●●●● ●●●●●●

●●

●●●●

●●●●●●

●●

●●●● ●

●●●

●●

●●●

●●●●

●●

●●●●

●●●●

● ●●●●●●

●●

●●

●●

●●●

● ●●●

● ●●●

●●

●●●●●●

●●●●●●●

●●

●●

●●●●●●

●●●●●●

●●

●●●

●●●

●●●

●●

●●

●●

● ●●●●

●●●

●●

●●

●●

●●●

●●●● ●

●●

●●

●●●●

●●●●

●●

●●

●● ●●●

●●●

●●

●●

● ●

Individual: 3

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●● ●

●●

●●

●●●

●●●

●●●

●●

●●

●●●

●●●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ● ●

● ●●

●●

● ●● ●

●●●●

●●●

●●

●●

●●

●● ●

●●●●

●●

●●

●●

● ●

●●

●●

●●

●●●

●●

●● ●

●●

●● ●●

● ●

●●

● ●●●

● ●

●●

●●

●●

●●●

●●●

●●

●●

●●●●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●● ●●● ●●●●●

●●

●●

●●●

●●●

●●

●●●

●●

●●

●●

●●●

●●●●

●●●

●●●

●●●

●●

●●●●

●●

●●●

●● ●

● ●

● ●●●

●●

●●●

● ● ● ● ●●● ●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ●

● ●●

● ●

●●

●●●

●●

●●

●●

● ●● ●

● ● ● ● ●●

●●

●●●

● ●

●●●●

●●●●● ●●● ●

●●

●●

●●

● ●●

● ●●●

●● ●●●●●

●● ●● ●

●●

● ●●

● ● ●

●●

●●

●●●

●●

●●

●●

●●●

●●●● ●● ●●

●●

●● ●

●●

●●

●●

●●

●●

●●●●

●●●●

●●

●● ●

●●

● ●●●

●● ●●

●●

●●

●●●

●●

●●

● ●

●●

●●

●●

●●

●●

●● ●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●●

● ● ●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

● ●●

●●

●●●● ●

●●

●●

●●

●● ●

●●

●●●

●●

●●

●●●

●●

●●

● ●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

● ●

●●

●●●

●●

●●●

●●

●●

●●●

● ●

●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●●

●●

● ●●

●●

●●●●

●●

● ●●

● ●●

●●

●●●

●●●

●●●

●●

●●

●●

●● ●●

●●

●●

●●●

●●

●●●

●●●●

● ● ●●

●●

●●

●●

● ●

●●

●●

●●●

●●

●●

● ●

●●

●●

●●

● ●●

● ●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

● ●

●●

●●

● ●●

●●

●●

●●

●●●

●●●●

●●

●● ●

●●●

● ●

●●

●●

●●●●

●●

●●

● ●

● ●●

● ●●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●●●●

●●

●●●

●●

●●●

● ● ●

●●

●●

●●

●●

●●●

●●

●●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

● ●

●●●

●● ●

●●●

●●

●●

●●

●●●

●●

●●

●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●●●

●●

●●

●● ●

●●

●●

●●

●●

● ●●

●●

●●●

● ●

●●

●●● ●●

●● ●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

● ●●

●●

●●

● ●●●

●●

●●

●● ●●

●●

●●

●●●

●● ●●

●●

●●●●

●●●

●●●

●●●●

●●●

●● ●●● ●●

●●● ●

●●●

●●

●●

●●●

●●

● ●

● ●●

●●

●●

●●

●●

●● ●● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

● ●

●●

●●●●

●●●● ●●

●●●

●●

●●

●●

●●

● ●

●●

●●

●●●●

● ●

●● ●

●●

●● ●

●●

●●

●●

●●●

●●

●●

●●

●●● ● ● ●

●●

●●

●●●

●●

●●●

● ●

●●●●

●●

●●●

●●●

●●

● ●

●● ●●

●● ●

● ● ● ●

●●

●●●● ●

● ●

●●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●●

●●

● ●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

● ●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●

● ●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ●●

● ●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●● ●

●●●

●●

●●

● ●●●

●●●

● ●

●●

●●

● ●●●●

●●

●●

●●●●●●●

●● ●●●●●●●●●●●●●●●●●

●●●●●●●

●●●●●●●●●●●●●●

●●

●●

●●

●●●●●●●

●●

●●●●

●● ●

●●

●● ●●●

●●●●

●●●

●●●●●

●●●●●●●●

●●

●●

●●

●●

●●

●●

●●●●●●

●●

●●●

●● ●●

● ●

●●

●●●

●●

●● ●

●●●●●●●

●●●●●●

● ●

●●●

●●●

●●

●●

●●●●

●●●●

●●

●●

● ●

●●●●●●●●●●

●●●●

●●●

●●●●

●●

●●●●●●●

●●●●●●

●●

●●●●●●●

●●

●●

●●●●●●

●●●●

●●●●●

●●●

●●

●●●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●●●●●●●●●

●●●●●●

●● ●●

●●●

●●

● ●●

●●

●●●●●●

●●

●●

●●

● ●●

●●

●●●●

●●●●●●

●●

●●●

●●

●● ●●●

●●●●● ●●

● ●

●●●

●●●

●●●

●●

●●●●

●●●

●●●

●●

● ●●●●●

●●

●●

●●●●●●●

●●

●●

●● ●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ● ●

●●●

● ●●●

●●

●●

● ●●

● ●●●

●●

●●●

●●

● ●●

●●

● ●

●●●●●

●●

●●

●●●

● ●●

●●●●●●

Individual: 4

●●

●●

●●●

●●

●●

●●

●●●●

●●●

●● ●●

●● ● ●

●●●

●● ●

●●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●

●●

●●

●●

● ●●●

●●●

●●

●●

●●

●●●● ● ●● ●

●●

●●●

●●●●●●●

●●

● ●

●●

●●●●

●●

●●

●●

● ●●

●●

●●

●●

●● ●●

●●

●●●

●●

●●●

●●

●●●

●●●

●●

●●

●●

●●

●●●

●●●●

●●● ●

●●

●●●

●●

● ●●●

●●

● ●

●●

●●

●●

●●

● ●

● ●●●●●

● ● ●

●●

●●

●●

●●

●●

● ● ● ●●

●●

●●

●●

●●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

● ● ●● ●● ●●●

●●

●●●

●●

●●

● ●●

●●● ●

●●●

● ●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●

● ●

●●

●●

●●

●●●●●●● ●

●●●●●●●●

●●

●●●

●●

● ●●

●●●

●●

●●

●●

●●●

●● ●

●●●●

●● ●

●●

●●

●●●●

●●●●

●●

●●

● ● ●●●●

● ●●●

●●

●●

● ●●

●●

●●

●●●

●●

● ●

●●

●●

●●

● ● ●●

●●

● ●●●

●●●

●●

●●

●● ●●●

● ●●●

●●

●●

●●●

● ●

●●

●●●

●●●

●●

●● ●●

●● ●●●●

●●●●●●●

●●

●●

●●

●●●

●●

●●●

●● ●●

●● ●

●●

●●● ●● ●

●● ●●

●●

●●

●●●● ●

● ●

●●

● ●

●●

●●

● ●●

●●●●

●●

●● ●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●●●●●● ●

●●

●●●●●

●●

●●●●● ●

●● ● ●●●●

●●●

●●● ●

●● ●● ●●●● ●●●●

●●

●● ●●

●●

●●●●

●●●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●●

●●●

●●●

●●● ●

●●

●●

●●

●●●●

●●

●● ●●

●●

●●

● ●●

●●

●●

●●

●●●●

●●●●●●

●●●

●●●

● ●

●●

●●

●●●

●●

●●●

● ●●●●

●●● ●● ●

●●●●●

●● ●

●●●

●●

●●

●●●●

● ●● ●●

●●

●●

●●

●●●

●●●

●●

● ●●●●●●● ●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

● ● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●● ●

● ●

●●

●●●

● ●●

●●

●●

●●

●● ●●

●●

●● ● ●

●●●● ●●●●

●●

●●

●●●●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●● ●

●●

●●

●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

● ●

●●

●●

●●●●

●●

●●

●●

●●

●● ●

●●●

●●●

● ●●

●●

●●

●●●

●●

●●

●●●●

●●●●●●

●●●

●●

●●

●●

●●

●●

● ●●

●●

●●●●

● ●●

●●●●●●●●

●● ●●●●●●●●

●●●●●●●●●●●●●●●●●●

●●●●●●●●●

●●

●●●●

●●●●●●●

●●●

●●●●●

●●

●●●●●●

● ●

●●●●●●●●

●●

●●

●●

●●

●●

●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●

● ●● ●

●●●

●● ●

●●

●●●

●●●●●●●●●●●

●●●●●●●●●

● ●

●●

●●●●

●●●●●●●●

●●

●●●●

●●●●●●●●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●●●

●●●

●●●●●●●●●

●●●●●●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●●●

●●●●●●●●●●●●●●●●●●●●

●●●●●●

●●●●●

●●●●●●●●●

●●●●●●●●●●●

●●●

●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●

●●●●●●●●●●●●●●●● ●●

●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●● ●●●●●●●●●●●●

●●●●●●●●●●●●●●●

●●●●●

●●●● ●

●●

●●●●

●●

●● ●●●●●●

●●●●●

●●●●●

●●●

● ●

●●

●●●●●

●●●●●● ●●●

●●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

● ●

●●●

●●●●

●●

●●

●●●●●

●●

Individual: 5

●●

● ●

● ●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●

●●●●●

●●

●●●

●●

●●

●●●

● ●●●

●●

●●

●●●● ●

●●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●● ●

●●

●●

●●

●●●

●●

●●

●●

●● ●●

● ●

●● ●

●●

●●●

●●●●●●

● ●●

●●

●●

●●

●● ●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

● ● ●

● ●●

●● ●

●●

●●

● ●

● ●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●●● ●

● ●●●

●●

●● ●

●●

●●

●●●

●●●

●●

●●

●●●

●●●●

●●●

● ●

● ●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

● ●

● ●

● ●

●●

●●

●●

●●

●●

●●● ● ●

●●

●●

●●

●●●

●●●

●●

● ●●

●● ●●

●●●

●●●●

●●

●●

● ●

●●

●●

●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

● ●●●

●●●●

● ●

●●●

● ●

●●●

●●●● ● ● ● ● ●

●●

●●

●●●● ●●●●

●●●

● ●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●

●●●

●●

●●●●●

●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●

●●

●●●

●●

● ●●● ● ● ● ●●

●●

● ●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

● ●

●●

●●

● ●

●●●

●●

●●

●●

●●●● ●●● ●●

●●

●●

●●●●●

●●

●● ●●

●●

●●

●●● ●

●●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●●●

●●

●●

●●

●●●

●●●

●●

●● ● ●●●●

●●

●●

● ●●●●

●●

●●

●●

●●

●●●●●●●

●●●●●

●●●●

●●● ● ●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●●

●●

● ● ● ●● ●

●●●

●●

●● ●●●

●●●

●●

●●

●●●

●●

●●●

●●

●●

●●●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●● ●

●●●

●●

●●

●● ●●

●●●

● ●

●●

●●

●●●

● ●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●●

●●●●

●●

●●

●●

●● ●

●●●

● ●●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●●

● ●

●●

●● ●●

●●

●●

●●

●●

●●

●●●●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●●

● ●●●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

● ●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●

●●●

●●●

● ●●

●●

●●

●●

●●

●●

●●

●●

● ●●● ●●

●●

●●

● ●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●●

●●

● ●

● ●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

● ●

●●●

●●

●●

●●

●●●●●

●●

●●

●● ●

●●

●●

●● ●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●● ●

●●

●● ●● ●●●

●●

●●

●●

● ●

●●

●●

● ●●

● ●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●●● ●

●●

●●●

●● ●

●●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

● ●

●●

●●● ● ●

●●

●●

●●●

●●●●

● ●

●●

●●

● ●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●● ●● ●● ●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

● ●

●●

●●

●● ●●● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

● ●●

●●●

●●

●●●

●●●

●●

●●

●●●●●●

●●●●

●●●

●● ●●

●●

●●

●●

●●●●

●●●●●

●●●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●●

●●●

●●

●●●●

●●

●●

●●

● ●

●●

●●

●●

● ●●

● ● ●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●● ●●●

●●

●●

●●

●●

●●

●●●

●●● ●

●●● ●● ● ● ● ● ●●

●●●●●●●●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●●

●●

●●

●●●●●

●●

●●●

●●●●●●

●● ●●

●●

●●●●●

●●●●

●●

● ●●

●●●●

● ●

●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●● ●

●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●● ●●●

●●

●●

● ●

● ●

●●

●●●●

●●

●● ●

●●●

●●

●● ●

●●

●●

●●

● ●●

●●

●● ●●●●●

● ●

●●

●● ● ●●

●● ●●

●●

●●

● ●

●● ●

●● ● ●

●●

●●

● ●

●●

●● ●

●●●●

● ●

●●

●●●

●●

●●

●●

● ●● ● ●

● ●● ●

● ●

●● ●

●●● ●●●

●● ●

●●

●●

●●

●●

● ●● ●

●●

● ●

●●

● ●●●●

●●

●●

●●

●●

●●

●●

●●●●●●

●● ●

● ●

● ●

●●

●●

●●

●●●

●●

● ●

● ●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●●

●● ●

●●

● ●●

●●

●●●●

●●

●●

●●●●

●●● ●● ●

● ●

● ●

●●

●●

●●

●●

●●

●● ●●

●●

●●

● ●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

● ●● ●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●●

●●●

●●

●●●

●●●●

●●

●●

●●

●● ●● ●

●●

●●

●●

●● ●

●●

●●

●●

● ●●●●

●●●

●●

● ●

●●

●●

●●

●●

●●●

● ●

●●●

●●●●

● ●● ●●

●●

● ● ●●

●●

●●

●● ●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●

● ●

● ●

●●

●●

●●

●●●● ●●

●●

●●

●●

●●●

●●●

●●

●●●

●●

● ●●

● ●

● ●

●●

●●

●●

●●

●●

●●

●● ●● ●

●● ●● ●●

●●

● ●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●● ●

● ●

●●

●●●

● ● ●

●●

●● ●

●●

●●●●

●●●

●●

●●

● ●

●●

●●

●●

● ●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

● ● ●●

● ●

● ●●

●● ●

●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

● ●●●●

●●●

●●

●●

● ●●

●●

● ●●●

●●

●●

●●

● ●●

●●

●●

●●

●●

●● ●●●

●●

● ●

● ● ●●●●●

●●

●●

●●

● ●●

●●

●●●

● ●

●●

●●

●●

●●●●

●●●

●●

●● ●

●●

● ●●● ●●

●●

● ●●

●● ●●

● ●

●●

●●

●●●

● ● ●

●●●

● ●

●●

●●

●●●

●●

●●

●●

●●● ●●●

●●

●●● ●●

●●●

●●

●●

●● ●●●

●●

●●

●●

●●●

●●

●● ●

● ●●● ●

●●

●●● ●

●●

●●

●●

●●

● ●●

●●

●●

●●

●●

● ●

●●

● ●

●●

●●

● ●●

●●

●● ●

●●

●●

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

Individual: 6

●●●●● ● ●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●●

●● ●

●●

●●

●●● ●

●●

●● ●

●●

●●●●

●●

●●

●●

●●

● ●

●●

● ●

●●●

●●

● ●● ●

●●●

●●

●●

●●

●●

● ●●

●●

●●

●●●

●●●●●

●●

●●

●●

●●●

●●●●

●●●●

●●

●●●

●●

● ●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●● ●●●●●

●●

●●

●●

● ●●

● ●●●

●●

● ●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●●

●●●● ●●●●●

●●

●●

● ●

●●

● ●●●

●●

●●

●●

●●

●●

●●

● ●●●

●●

● ●

●●

●●

●●●

●●●

●● ● ●

●●

● ●●●

● ●●● ●●

●●

● ●●

●● ●

●●

●●●

●●●

●● ●

●●●●

● ● ●●

● ●●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

●●●

●●

●●●

●●●●

●●●● ●●

●●●●●

●●●

●●

●● ●

●●●

● ●

● ●● ●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●● ●●

●●●

●●

●●

●●

●●●●●●●●●●●●●

●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●●●●

●●●●

● ● ●●●

●●●●●

●●

●●●●●

●●

●●

●●●●

●●

●●

●●

●●●●●

●●

●● ●● ●●

● ●● ●● ●● ●●●●● ● ● ●

●● ●

●●

●● ●

●●

●●

●●

●●●

●●

●●

●●

●●

● ●

●● ●

●●

●●

●●

●●

●●

●● ●

●●●● ●●●

●●

●●●

●● ●●

●●

●●

●●●

●●

●●

●●

●●● ●

●●●

●●

● ●

●●

●●

●●●●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●● ●

●●

●● ●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

● ●●

●●

●●

●● ●

●●●

●●

●●

●● ●

●●

●●●

●●

●●● ●

●●

●●●●

●●●●●

●●

●●●

●●

● ●

●●

● ●

● ●

●●

●●

●●●

●●● ●

●●●●

●● ●

●● ●

●●

●●

● ● ●●

●●

●● ●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

● ● ●●

●● ●

●●

●●

●●

● ●●

●●

●●

●●●

● ●

●●

●●

●●

●●●

●●

●●

● ●

●●

●● ●●● ●

●●

● ●●

●● ●

●●●

●●

●● ●●

●●

● ●●● ●●

●●

●●

● ●●●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

● ●●

● ●●

●●

●●●

●●

●●●

●●●

●●

●●●

●●

●●

●●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●● ●●●

●●

● ●●

●●

●●

●●

●●

●● ●●●●● ●● ●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●●

●●

●●

●●●

●●

● ● ●

●●

●●

●●● ●

●●

● ●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●● ●

●● ●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●● ● ●

●●

●● ●●

●●

●● ●

●●● ●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●●●

●●

●●● ●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●● ●●

●●

●●

●●

●●

●●

●●●●●●

●●

●●

●● ●●

● ●

●●

●●

● ●

●●

●●●●● ●

●● ●

●●●

● ●

●●

● ●

●●

●●● ●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●● ●●

●●

●●

●●●

●●

●●

●● ●● ●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

● ●

●●

●●

● ●●

●●●

●●●●

●●●

●●

●●

●●

●● ●●●●

●●

●●

●●●

●●●●

●●

● ●

● ●

●●●

●●

●●

●●

●●

●●

●●●●

●●●

●●●

●●●●●

●●

●●●

● ●

●●

●●●

●●●

●●

●●

● ●

●●

●●

●●

● ●●

● ●●●

●●●

●●●

● ●

●●● ●

●● ●

●●●

● ●

●●

●●

●●● ●

●●

●●

●●

● ●

●●● ●● ●

●●

●●●

●●●

●●

●●

●●

●●

●●●● ●●

●●

● ●

●●●

●●

●●

●●

●●

● ●●●●

Individual: 7

●●●

●●

●● ●

●●●

●●●● ●● ●

●● ●●

● ●● ●●

●●

●●●●●●

●●●● ● ●● ● ●

●● ●

● ●●

●●● ●

●●

●●●

●●

●●

●●

●●●●

●● ●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●

●●

●●

●●

●●●

●●●●●●

●●

●●

●●

●●●

●● ●

●●

● ● ●●

●●

●●●

●●

●●

●●●●

●●●

●●●

●●

●●

●●

●●

● ●●●

●●●

●●

●●

●●●●

●●●

●●

●●

●●●

●●

●●●

●●

●● ●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●●

●●

●●●●

●●

●●

●●

●●

● ●

●●

●●●●●

● ●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●● ●

●●●

●●

●●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●● ●

●●

●●

●●

●●

● ●

●●●

●●●

●●●

●●●●

● ●●●

●●●

●●

●●●

●●●●●

●●

● ●●

●●●

●●

●●

●●

●●●●

●●●

● ●

● ●●● ●

●●

●●

●●

●●

● ●

● ●

●●

●●

●●

●●●

●● ●● ●

●● ●●

●●

●● ●

●●●

●● ●

● ●

●●

● ●

●●●

●●●

●●

●●●● ●

●● ●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

● ●

●●

● ●

●●

●●

●●

●●

● ●

●●●

●●

● ●

●●

●●

●●

●●

●●●

●●

●●●

● ●

●●

●●

●●●

●●

●●●

●●

● ●

●●

●●

●●

●●

●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●

●●

●●●

● ●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

● ●●

●●

●●

●●●

● ●●

●● ●●

●●●●

●●

●●●

●●● ●●

●●

●●

● ●●

●●●●

●● ●

●●

●●

●●●

●●

● ● ●●●●●

Individual: 8

19 / 28

Page 49: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Application of Density Ranking: GPS dataset - 2

Individual: 1 Individual: 2 Individual: 3 Individual: 4

Individual: 5 Individual: 6 Individual: 7 Individual: 8

20 / 28

Page 50: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summarizing Multiple Density Ranking: Level Plots

◦ In the above example, we have multiple GPS datasets that lead tomultiple density ranking.

◦ To compare these density rankings, a simple approach is tooverlap level plots.

◦ For a density ranking α, let

Aγ � {x : α(x) ≥ 1 − γ}be the (upper) level set.

◦ We can compare the density ranking of each individual byoverlapping their level sets at each level.

21 / 28

Page 51: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summarizing Multiple Density Ranking: Level Plots

◦ In the above example, we have multiple GPS datasets that lead tomultiple density ranking.

◦ To compare these density rankings, a simple approach is tooverlap level plots.

◦ For a density ranking α, let

Aγ � {x : α(x) ≥ 1 − γ}be the (upper) level set.

◦ We can compare the density ranking of each individual byoverlapping their level sets at each level.

21 / 28

Page 52: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summarizing Multiple Density Ranking: Level Plots

◦ In the above example, we have multiple GPS datasets that lead tomultiple density ranking.

◦ To compare these density rankings, a simple approach is tooverlap level plots.

◦ For a density ranking α, let

Aγ � {x : α(x) ≥ 1 − γ}be the (upper) level set.

◦ We can compare the density ranking of each individual byoverlapping their level sets at each level.

21 / 28

Page 53: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.1

22 / 28

Page 54: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.2

22 / 28

Page 55: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.3

22 / 28

Page 56: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.4

22 / 28

Page 57: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.5

22 / 28

Page 58: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.6

22 / 28

Page 59: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.7

22 / 28

Page 60: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.8

22 / 28

Page 61: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.9

22 / 28

Page 62: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Level Plots: Example

γ=0.9

22 / 28

Page 63: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summary Curves of Density Ranking

◦ The level plot allows us to compare GPS datasets from differentindividuals.

◦ However, it has two drawbacks:◦ When we have more individuals, this approach might not work (too

many contours).◦ We often need to choose a level γ to show the plot but which level

to be chosen is unclear.

◦ Here we introduce a few curves to summarize geometric andtopological features of density ranking.

23 / 28

Page 64: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summary Curves of Density Ranking

◦ The level plot allows us to compare GPS datasets from differentindividuals.

◦ However, it has two drawbacks:◦ When we have more individuals, this approach might not work (too

many contours).◦ We often need to choose a level γ to show the plot but which level

to be chosen is unclear.

◦ Here we introduce a few curves to summarize geometric andtopological features of density ranking.

23 / 28

Page 65: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Summary Curves of Density Ranking

◦ The level plot allows us to compare GPS datasets from differentindividuals.

◦ However, it has two drawbacks:◦ When we have more individuals, this approach might not work (too

many contours).◦ We often need to choose a level γ to show the plot but which level

to be chosen is unclear.

◦ Here we introduce a few curves to summarize geometric andtopological features of density ranking.

23 / 28

Page 66: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Mass-Volume Curve

◦ Recall that Aγ � {x : α(x) ≥ 1 − γ} is the level set of densityranking.

◦ The mass-volume curve is a curve of(γ,Vol(Aγ)

): γ ∈ [0, 1].

◦ Namely, we are plotting the size of set Aγ at various level.

◦ In practice, we often plot γ versus log Vol(α)γ.

24 / 28

Page 67: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Mass-Volume Curve

◦ Recall that Aγ � {x : α(x) ≥ 1 − γ} is the level set of densityranking.

◦ The mass-volume curve is a curve of(γ,Vol(Aγ)

): γ ∈ [0, 1].

◦ Namely, we are plotting the size of set Aγ at various level.

◦ In practice, we often plot γ versus log Vol(α)γ.

24 / 28

Page 68: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Mass-Volume Curve

◦ Recall that Aγ � {x : α(x) ≥ 1 − γ} is the level set of densityranking.

◦ The mass-volume curve is a curve of(γ,Vol(Aγ)

): γ ∈ [0, 1].

◦ Namely, we are plotting the size of set Aγ at various level.

◦ In practice, we often plot γ versus log Vol(α)γ.

24 / 28

Page 69: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Mass-Volume Curve: Example

0.0 0.2 0.4 0.6 0.8 1.0

−3

−2

−1

01

23

Mass−Volume Curve

γ

log(

km2 )

12345678910

25 / 28

Page 70: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Betti Number Curve

◦ The Betti number curve is a curve quantifying topological featuresof the density ranking.

◦ It counts the number of connected components of Aγ at variouslevel γ.

◦ Formally, the Betti number curve is(γ, Betti0(Aγ)

): γ ∈ [0, 1],

where for a set A

Betti0(A) � number of connected components inside A.

◦ Note that the number of connected component is called the 0thorder Betti number (0th order topological structure); one cangeneralize this idea to higher order topological structures.

26 / 28

Page 71: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Betti Number Curve

◦ The Betti number curve is a curve quantifying topological featuresof the density ranking.

◦ It counts the number of connected components of Aγ at variouslevel γ.

◦ Formally, the Betti number curve is(γ, Betti0(Aγ)

): γ ∈ [0, 1],

where for a set A

Betti0(A) � number of connected components inside A.

◦ Note that the number of connected component is called the 0thorder Betti number (0th order topological structure); one cangeneralize this idea to higher order topological structures.

26 / 28

Page 72: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Betti Number Curve

◦ The Betti number curve is a curve quantifying topological featuresof the density ranking.

◦ It counts the number of connected components of Aγ at variouslevel γ.

◦ Formally, the Betti number curve is(γ, Betti0(Aγ)

): γ ∈ [0, 1],

where for a set A

Betti0(A) � number of connected components inside A.

◦ Note that the number of connected component is called the 0thorder Betti number (0th order topological structure); one cangeneralize this idea to higher order topological structures.

26 / 28

Page 73: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Betti Number Curve: Example

0.0 0.2 0.4 0.6 0.8 1.0

050

100

150

Betti Number Curve

Num

ber

of C

onne

cted

Com

pone

nts 1

2345678910

27 / 28

Page 74: Geometric and Topological Data Analysisfaculty.washington.edu/yenchic/Talks/201709Intro.pdfGeometric and TopologicalDataAnalysis: Big Picture Thedatacanbeviewedas X 1; ;Xn ˘p; p isaprobabilitydensityfunction

Density Ranking: Open Questions

◦ Convergence of density ranking level sets.

◦ Convergence of summary curves under singular/non-singularmeasure.

◦ Other summary curves.

◦ Convergence of higher order topological structures.

◦ Connection to stratified space.

28 / 28