Contents

• 1 Introduction
  • 1. Introduction
    • 1.1 Science In This Century
    • 1.2 Focus Of Thesis
    • 1.3 Organization of Thesis
  
  
• 2 Background
  • 2. Conjugate Gradient
    • 2.1 Minimizing Quadratic Forms
      • 2.1.1 The Wrong Iterative Method: Steepest Descent
      • 2.1.2 The Right Iterative Method: Conjugate Gradient
    • 2.2 Minimizing Arbitrary Convex Functions With Conjugate Gradient
  • 3. Linear Regression
    • 3.1 Normal Model
    • 3.2 Ridge Regression
  • 4. Logistic Regression
    • 4.1 Logistic Model
    • 4.2 Maximum Likelihood Estimation
    • 4.3 Iteratively Re-weighted Least Squares
    • 4.4 Logistic Regression for Classification
  
  
• 3 Logistic Regression For Fast Classification
  • 5. Logistic Regression Computation
    • 5.1 Preliminaries
      • 5.1.1 Why LR?
      • 5.1.2 Our Approach
      • 5.1.3 Datasets
      • 5.1.4 Scoring
      • 5.1.5 Computing Platform
      • 5.1.6 Scope
    • 5.2 IRLS Parameter Evaluation and Elimination
      • 5.2.1 Indirect (IRLS) Stability
      • 5.2.2 Indirect (IRLS) Termination And Optimality
      • 5.2.3 Indirect (IRLS) Speed
      • 5.2.4 Indirect (IRLS) Summary
    • 5.3 CG-MLE Parameter Evaluation and Elimination
      • 5.3.1 Direct (CG-MLE) Stability
      • 5.3.2 Direct (CG-MLE) Termination And Optimality
      • 5.3.3 Direct (CG-MLE) Speed
      • 5.3.4 BFGS-MLE
      • 5.3.5 Direct (CG-MLE) Summary
  • 6. Characterizing Logistic Regression
    • 6.1 Classifiers
      • 6.1.1 SVM
      • 6.1.2 KNN
      • 6.1.3 BC
    • 6.2 Synthetic Datasets
      • 6.2.1 Number of Rows
      • 6.2.2 Number of Attributes
      • 6.2.3 Sparsity
      • 6.2.4 Attribute Coupling
    • 6.3 Real-world Datasets
    • 6.4 Conclusions
  
  
• 4 Conclusion
  • 7. Related Work
  • 8. Conclusions
  • 9. Contributions
  • 10. Future Work
  • A. Acknowledgements
  • B. Software Documentation
    • B.1 Introduction
    • B.2 Actions
      • B.2.1 Action Summary
      • B.2.2 Action Details
    • B.3 Data
      • B.3.1 Dataset files
      • B.3.2 Sparse Datasets
      • B.3.3 Dataset naming
    • B.4 Learners
      • B.4.1 bc
      • B.4.2 dtree
      • B.4.3 lr
      • B.4.4 lr_cgmle
      • B.4.5 newknn
      • B.4.6 oldknn
      • B.4.7 super
      • B.4.8 svm
    • B.5 Examples
      • B.5.1 Simple roc examples on several datasets
      • B.5.2 Examples with various actions
      • B.5.3 Examples with various learners
  • C. Miscellanea
  
  
• Bibliography
• Index