본문 바로가기

Machine Learning

 (5)
Clustering 1. Unsupervised learning introduction 2. K-means algorithm 3. Optimization objective 4. Random initialization 5. Choosing the number of clusters Unsupervised learning introduction Supervised learning Training set: {(x₁, y₁), (x₂, y₂), (x₃, y₃), ... } Unsupervised learning Training set: {x₁, x₂, x₃, ... } Clustering 적용 분야: Market segmentation, Social network analysis, Organize computing clusters,..
Dimension Reduction 1. Principal Component Analysis 2. Multidimensional Scaling Principal Component Analysis Principal Component Analysis (PCA) variation(변동성)이 가장 큰 axis를 찾아 projection 하여 dimension reduction Reduced 된 dimension은 모두 orthogonal Eigen-decomposition Algorithm Multidimensional Scaling Multidimensional Scaling (MDS) dimension reduction 전·후 좌표값들의 pairwise distance를 최대한 보존하는 게 목적 → distanced value값을 측정하여 최..
Cross Validation & Dimension Reduction 1. Cross Validation 2. Dimension Reduction Cross Validation Leave-One-out Cross Validation (LOOCV) 모든 data에 대하여 한 번씩 validation 수행 성능은 K-fold CV 보다 좋으나, 수행 시간이 오래 걸림 K-fold Cross Validation (k-fold CV) data를 K개의 fold로 균등하게 나누어 validation 수행 성능은 LOOCV보다 떨어지나, 수행 시간이 덜 걸림 Data를 Train, Validation, Test data로 분할 → Cross Validation 적용 Dimension Reduction Dimension Reduction Data의 dimension을 축소 memory..
Regularization 1. Overfitting 2. Regularization Overfitting Underfitting: High bias, Low variance Overfitting: Low bias, High variance → 새로운 sample에 대한 일반화 ↓ overfitting 감소법 1. Feature 개수 감소 적절한 Feature들만 선택 Model selection algorthm 2. Regularization 모든 Feature들 유지한 채, parameters θj의 value 감소 Feature들 기여도 감소 Regularization Parameter λ↑ Regularization paramter λ↓ Feature 수↑ Feature 수↓ go Underfitting go Overfit..
Logistic Regression 1. Classification 2. Hypothesis Representation 3. Decision boundary 4. Cost function 5. Simplified cost function and gradient descent 6. One-vs-all Classification Email: Spam / Not Spam ? Online: Transactions: Fraudulent (Yes / No) ? Tumor: Malignant / Bening ? → y ∈ {0, 1} 0: "Negative Class" 1: "Positive Class" Hθ(x) = θ^Tx Hθ(x) ≥ 0.5, predict "y=1" Hθ(x) < 0.5, predcit "y=0" Classification: ..

티스토리툴바