CSCI2750
Data Mining and Information Processing
Clustering, frequent pattern mining, classifiers from naive Bayes to support vector machines, classifier evaluation, principal components, sampling and image histograms.
Your progress
0/ 20 sections marked complete
Next section: Means, objectives and k-means clustering
Completion marks are saved in this browser.
This course in the web
In the CSCI2750 notes
- Theorem cards
- 39
- Definitions
- 30
- Counterexamples
- 25
- Concept lenses
- 11
- Histories of ideas
- 1
- Proof X-rays
- 20
Sections with a history of an idea
None of these is on the timeline yet.
No concept lens from this course is recorded in the web yet.
Counts cover the 20 public sections of this course and what is written in them, not recorded connections. 0 of these sections are indexed into the web in detail so far.
In the webTheorem cards 39 · Definitions 30 · Counterexamples 25 · Concept lenses 11 · Histories of ideas 1 · Proof X-rays 20
None of these is on the timeline yet.
No concept lens from this course is recorded in the web yet.
Counts cover the 20 public sections of this course and what is written in them, not recorded connections. 0 of these sections are indexed into the web in detail so far.
Chapter 1 · 1 section
Centroid clustering
Centroid clustering
Theorems 2 · Definitions 1Counterexamples 2Lenses 1Proof X-rays 2
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 1.1
Means, objectives and k-means clustering
Derive the centroid update, trace exact assignment and update steps, and distinguish a stable clustering from a globally best one.
Chapter 2 · 2 sections
Pattern mining
Pattern mining
Theorems 7 · Definitions 3Counterexamples 3Lenses 1History 1Proof X-rays 2
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 2.1
Support counts, safe pruning and Apriori
Derive the support property behind Apriori, trace candidate generation and pruning, then interpret association rules without confusing their measures.
- 2.2
Frequent, closed and maximal itemsets
Compare support-preserving closed itemsets with maximal frequent itemsets and recover support counts.
Chapter 3 · 4 sections
Classification
Model-based classification and decision trees.
Theorems 8 · Definitions 6Counterexamples 4Lenses 2Proof X-rays 3
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 3.1
Naive Bayes and smoothing
Build binary-feature class scores, normalize them, and examine the independence assumption.
- 3.2
Decision trees: weighted impurity, thresholds and generalization
Prove majority-leaf optimality and nonnegative information gain, work correctly weighted splits, handle repeated feature values, and distinguish greedy training gains from predictive performance.
- 3.3
Bernoulli naive Bayes: log odds and decisions
Extend smoothed Bernoulli classification with log odds, cost-sensitive decisions, absent features and dependence counterexamples.
- 3.4
k-Nearest-Neighbor Classification
Classify by a vote among the k closest training examples, with explicit tie rules, feature scaling, leakage-free preprocessing and prediction cost.
Chapter 4 · 1 section
Sampling
Understand discrete samples and aliasing.
Theorems 1 · Definitions 1Counterexamples 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 4.1
Sampling, aliasing and frequency ambiguity
Derive why distinct sinusoids can produce identical discrete samples.
Chapter 5 · 1 section
Perceptron updates
Strict margins, update rules and a finite-update bound.
Theorems 1Counterexamples 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 5.1
The perceptron: margins and a finite-update bound
Prove a finite-update guarantee under a strict separation assumption.
Chapter 6 · 1 section
Logistic models
Binary likelihood, convex loss and separable data.
Theorems 2 · Definitions 1Counterexamples 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 6.1
Logistic loss, separability and regularization
Derive logistic loss and show how separability and a penalty affect attainment.
Chapter 7 · 2 sections
Classifier evaluation
Precision, recall and class prevalence.
Theorems 2 · Definitions 2Counterexamples 3Proof X-rays 2
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 7.1
Precision, recall and changing base rates
Interpret confusion counts and derive precision from conditional rates and prevalence.
- 7.2
Reading ROC curves without losing the ties
Build a ROC curve from scores with tied values, compute AUC by trapezoids and by pairwise comparison, and relate ROC to precision and decision costs.
Chapter 8 · 1 section
Intensity histograms
Construct a discrete cumulative histogram and intensity mapping.
Theorems 1 · Definitions 1Counterexamples 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 8.1
Image histograms and cumulative intensity mappings
Compute a discrete cumulative histogram and a global intensity mapping.
Chapter 9 · 1 section
Linear classification practice
Revisit the perceptron margin bound, trace affine updates and verify the outcome of a bounded training run.
Theorems 1 · Definitions 2Counterexamples 2Lenses 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 9.1
Affine perceptron updates and stopping
Revisit the perceptron margin bound, trace affine updates and verify the outcome of a bounded training run.
Chapter 10 · 1 section
Agglomerative clustering
Linkage updates, dendrograms and threshold connectivity.
Theorems 4 · Definitions 3Counterexamples 2Proof X-rays 2
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 10.1
Agglomerative clustering: linkage, dendrograms and threshold connectivity
Derive single, complete and average linkage and interpret a single-linkage cut.
Chapter 11 · 1 section
Support vector machines
Margins, hinge loss, dual certificates and kernels.
Theorems 3 · Definitions 2Counterexamples 1Lenses 2Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 11.1
Margins, hinge loss and the meaning of a kernel
Separate scores from geometric margins, derive the hard- and soft-margin objectives with a dual certificate, and decide when a function is a valid kernel.
Chapter 12 · 1 section
Source separation
Linear mixtures, whitening and independent component analysis.
Theorems 3 · Definitions 2Counterexamples 1Lenses 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 12.1
Blind source separation: what mixtures can and cannot reveal
Model sensor readings as a linear mixture, show why blind separation is ambiguous, whiten with the covariance matrix, and separate uncorrelated from independent.
Chapter 13 · 1 section
Principal component analysis
Principal directions as eigenvectors of the sample covariance, with variance and reconstruction guarantees.
Theorems 1 · Definitions 2Counterexamples 1Lenses 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 13.1
Principal component analysis: centered covariance and variance maximization
Define principal components as eigenvectors of the sample covariance matrix, prove the variance-maximizing and reconstruction properties, and compute explained variance on exact examples.
Chapter 14 · 1 section
Distances and data preprocessing
Metrics, similarity measures and feature scaling without test-data leakage.
Theorems 2 · Definitions 3Counterexamples 1Lenses 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 14.1
Distance and similarity measures, and feature scaling
Define metrics and the Minkowski, cosine, Jaccard and Hamming measures, prove which satisfy the triangle inequality, and apply min-max and z-score scaling without leaking test data.
Chapter 15 · 1 section
Density-based clustering
Core, border and noise points, density-reachability and the DBSCAN algorithm.
Theorems 1 · Definitions 1Counterexamples 1Lenses 1Proof X-rays 1
Chapter outline →See this chapter on the web →Not yet indexed in the web
- 15.1
DBSCAN: core, border and noise points
Define density-reachability, state the DBSCAN algorithm, prove that clusters of core points are well defined, and run it on exact one-dimensional data with a parameter study.