構造的表現 Structure and Space

Slides:



Advertisements
Similar presentations
図示、可視化モジュール ~ pylab と numpy を ちょっと~. pylab とは? ・数学や統計的なグラフを生成するモ ジュール ・インストール pip や easy install からのインストールを推奨 →numpy モジュールなどの前提としている。 Anaconda の場合は標準.
Advertisements

第 5 章 2 次元モデル Chapter 5 2-dimensional model. Contents 1.2 次元モデル 2-dimensional model 2. 弱形式 Weak form 3.FEM 近似 FEM approximation 4. まとめ Summary.
ロジスティクス工学 第6章 動的ロットサイズ決定モデル 東京商船大学 久保 幹雄
Cluster Analysis 徳山研究室M2 鈴木 晶子.
人工知能特論 8.教師あり学習と教師なし学習
第8章 ベクトル・行列の基礎 Rによるベクトル・行列の表現と計算法.
How to Measure Genetic Heterogeneity
On the Enumeration of Colored Trees
2010年7月9日 統計数理研究所 オープンハウス 確率モデル推定パラメータ値を用いた市場木材価格の期間構造変化の探求 Searching for Structural Change in Market-Based Log Price with Regard to the Estimated Parameters.
Extremal Combinatorics 14.1 ~ 14.2
なぜ 今の どうする どうやって なにを どのように だれが だれと (変える) どこで だれの 考え方9W4H(Ver.0) いつ から
What did you do, mate? Plain-Past
先端論文紹介ゼミ Role-based Context-specific Multiagent Q-learning
SP0 check.
Probabilistic Method 6-3,4
Shell model study of p-shell X hypernuclei (12XBe)
Licensing information
3次元での回転表示について.
What is the English Lounge?
クラスタリング 距離と分類の考え方.
Kalman Filter Finite Element Method Applied to Dynamic Motion of Ground Yusuke KATO Department of Civil Engineering, Chuo University.
フードピラミッド(Food Pyramid)
ソースコードの変更履歴における メトリクス値の変化を用いた ソフトウェアの特性分析
ロジスティクス工学 第7章 配送計画モデル 東京商船大学 久保 幹雄
分割表 Contingency table.
量的表現 Quantitation.
データベース工学 生研 戦略情報融合研究センタ 喜連川 優.
Probabilistic Method 2007/11/02 岩間研究室M1 市場孝之.
IIR輪講復習 #17 Hierarchical clustering
Where is Wumpus Propositional logic (cont…) Reasoning where is wumpus
大規模なこと Large scale.
SksMinus status 23 HB meeting 2009/3/19 白鳥昂太郎.
3次元での回転表示について.
Anja von Heydebreck et al. 発表:上嶋裕樹
多変量解析ゼミ 第10回 第12章クラスター分析 発表者 直江 宗紀.
東北大学 大学院情報科学研究科 応用情報科学専攻 田中 和之(Kazuyuki Tanaka)
Question Words….
G. Hanson et al. Phys. Rev. Lett. 35 (1975) 1609
遺伝統計学の骨組み Skeleton of Genetic Statistics
2019年4月8日星期一 I. EPL 84, (2008) 2019年4月8日星期一.
Statistical Genetics 統計遺伝学
X軸方向にa間隔、Y軸方向にb間隔で並んだ格子点 (単位格子:a×bの長方形) ミラー指数(2次元の例) a
Data Clustering: A Review

Data Clustering: A Review
出典:小島尚人、舘智士:インターネット環境下で稼働する画像幾何学的歪補正システムの構築、土木情報技術論文集、Vol 出典:小島尚人、舘智士:インターネット環境下で稼働する画像幾何学的歪補正システムの構築、土木情報技術論文集、Vol.18、pp249~260、2009年10月.
Statistical Genetics 統計遺伝学
Statistical Genetics 統計遺伝学
1-3 UMLの図(ダイアグラム) コンポーネント図 システムの物理的な構成を表現 ソフトウェアコンポーネントの依存性を表現
Genetic Statistics Lectures (4) Evaluation of a region with SNPs
構造的表現 Structure and Space
先進的データ分析法 Advanced Data Analysis
北大MMCセミナー 第62回 附属社会創造数学センター主催 Date: 2016年11月4日(金) 16:30~18:00
遺伝関係のグラフ Hereditary graphs
千代浩司 高エネルギー加速器研究機構 素粒子原子核研究所
千代浩司 高エネルギー加速器研究機構 素粒子原子核研究所
行列式 方程式の解 Cramerの公式 余因数展開.
パターン認識 ークラスタリングとEMアルゴリズムー 担当:和田 俊和 部屋 A513
パターン認識 ークラスタリングとEMアルゴリズムー 担当:和田 俊和 部屋 A513
ベイズ音声合成における 事前分布とモデル構造の話者間共有
Visualizing Japanese Grammar Appendix
Grammar Point 2: Describing the locations of objects
制約付き非負行列因子分解を用いた 音声特徴抽出の検討
分割表 Contingency table.
Data Clustering: A Review
重力波解析ライブラリKagaliのためのAntenna pattern functionの作成
ガウシアングラフィカルモデルにおける一般化された確率伝搬法
グラフ-ベクトル変換を用いたグラフ構造表現による一般物体認識
Improving Strategic Play in Shogi by Using Move Sequence Trees
千代浩司 高エネルギー加速器研究機構 素粒子原子核研究所
Presentation transcript:

構造的表現 Structure and Space

似ている 似ていない Alike, not alike

実数直線 (Real) number line

2項目で距離を定める

k項目で距離を定める

距離 Distance L(A,B) L(A,B)>=0 L(A,B) = L(B,A) L(A,A) = 0 A value is defined between two items, A and B L(A,B)>=0 L(A,B) = L(B,A) L(A,A) = 0 L(A,B) <= L(A,C) + L(B,C) 三角不等式 Triangle inequality

三角不等式 Triangle inequality 次を解け Solve a=y+z; b=z+x; c=x+y.

ペアワイズ Pair-wise L(A,B) N*(N-1)/2 2つのものA,Bの間に定められる量 A value is defined between two items, A and B N*(N-1)/2

近隣結合法 Neighbor-Joining method 三角不等式を用いて木を描く Draw a tree using triangle inequality

NJ法 NJ method #サンプル数Ns=5が5次元のデータを持っているようなデータセットを作成 R4-2.R #サンプル数Ns=5が5次元のデータを持っているようなデータセットを作成 Ns<-5;k<-5;x <- matrix(rnorm(Ns*k), nrow=Ns) dist(x,method="euclidean") # ユークリッド距離の距離行列 library(ape) # 木を作る関数nj()を持つパッケージapeの読み込み treu<-nj(dist(x,method="euclidean")) # apeの近隣結合法関数njにより木を作る par(mfcol=c(2,3)) plot(treu) # 木の表示 plot(treu, type="p") plot(treu, "c") plot(treu, "f") plot(treu, "u") plot(treu, "r") # type "phylogram”, "cladogram", "fan", "unrooted", "radial"

距離のとりかたを変える Different “distance-definitions” treu<-nj(dist(x,method="euclidean")) # apeの近隣結合法関数njにより木を作る trman<-nj(dist(x,method="manhattan")) # マンハッタン距離で木を作る par(mfcol=c(1,2)) # 画面を1行2列に分割 plot(treu);plot(trman) # 2つの距離法で木の表示 par(mfcol=c(1,1)) # 画面分割を1行1列に戻す

別のやり方で木を作る Different methods to generate trees R5-1.R 階層的クラスタリング法 Hierarchical clustering methods

距離行列の2軸 Two axes in distance matrix

行の順序をそのままに The order of rows as they are LD-plot R5-2.R R5-3.R m<-matrix(rbinom(120,1,0.5),20,6) heatmap(m) cormatrix<-cor(m);rsqmatrix<-cormatrix^2 image(1:nrow(rsqmatrix),1:ncol(rsqmatrix),rsqmatrix,col=gray((100:0)/100))

距離 角度 余弦 Distance Angle Cosine

ペアワイズ Pair-wise L(A,B) N*(N-1)/2 2つのものA,Bの間に定められる量 A value is defined between two items, A and B N*(N-1)/2

ペアワイズでない比較 Non-pairwise comparison 3要素の比較 Three items to be compared ABC 0 0 0 N1 0 0 1 N2 0 1 0 N3 0 1 1 N4 1 0 0 N5 1 0 1 N6 1 1 0 N7 1 1 1 N8 AB 0 0 N1+N2 0 1 N3+N4 1 0 N5+N6 1 1 N7+N8 BC 0 0 N1+N5 0 1 N2+N6 1 0 N3+N7 1 1 N4+N8 CA 0 0 N1+N3 0 1 N2+N4 1 0 N5+N7 1 1 N6+N8

ペアワイズでない比較 Non-pairwise comparison 3要素の比較 Three items to be compared ABC 0 0 0 0.125 0 0 1 0 1 0 0 1 1 1 0 0 1 0 1 1 1 0 1 1 1 ABC 0 0 0 0.25 0 0 1 0 1 0 0 1 1 0. 25 1 0 0 1 0 1 1 1 0 1 1 1 AB 0 0 0.25 0 1 1 0 1 1 BC 0 0 0.25 0 1 1 0 1 1 CA 0 0 0.25 0 1 1 0 1 1

関係はいくつある How many “RELATIONS”?

非階層的クラスタリング法 NON -hierarchical clustering methods R6-2.R 非階層的クラスタリング法 NON -hierarchical clustering methods R6-3.R

R6-2.R R6-3.R # 正規乱数を用いてデータを作る n1<-500;n2<-300;n3<-200;x<-c(rnorm(n1,0,0.5),rnorm(n2,5,1),rnorm(n3,8,2));y<-c(rnorm(n1,0,2),rnorm(n2,3,2),rnorm(n3,-3,1)) library(gregmisc) # hist2d()を持つパッケージ h2d <- hist2d(x,y, show=FALSE,same.scale=TRUE, nbins=c(10,10)) # 2次元ヒストグラム情報を取る plot(x,y) # 散布図 filled.contour( h2d$x, h2d$y, h2d$counts, nlevels=9,col=gray((8:0)/8) ) # 2次元ヒストグラムを濃淡で persp( h2d$x, h2d$y, h2d$counts,ticktype="detailed", theta=60, phi=30,shade=0.5, col="cyan") # 2次元ヒストグラムを鳥瞰図で # 非階層的クラスタリング m3<-matrix(c(x,y),ncol=2) cl <- kmeans(m3, 3) # kmeans法で3群にクラスタリング plot(m3, col = cl$cluster) points(cl$centers,pch = 8, cex=10) # クラスタの中央に印をつける