Data Mining

Boosting
fit the model, weight the one heavily wrong, test the model again then repeat.

AdaBoost - a version of the AdaBoost algorithm
g is our "weak learner"
classes -1 & 1

correct prediction gives 1
incorrect prediction gives -1

Clustering-divide data into group
descriptive statistic
Goal- object within cluster is similar with each other, being closed together

-Clustering for understanding
-Clustering for Ultility
-Summaeriziing: different algorithms can run faster on a data set summarized by clustering
-Compression: big set of data, break into cluster, what are the key in each cluster
-Finding nearest neighbors

K-mean = number of cluster we make
each cluster associate with the center, often the mean
each point assign to cluster with the closest centroid
the number of cluster, K, must be specified ahead of time.