In order to solve the task T, design a program ,, e learning from experience, the performance reached p, if and only if the experience with E, after a judge P, program performance has been improved in dealing with T 
machine learning and human Similarly, according to history data model as a training experience,
there are labels, such as representatives supervised the outcome of red green (regression classification)
without labels, on behalf of unsupervised (clustering based on distance, split)
Classification, regression, clustering, time series analysis
Concept:
Characteristics: also called dimensions,
continuously variable :( numeric variables): Dimensions height and weight (generally regression)
discrete data: season, gender (usually to be bin cut, qcut, or normalized) (generally with classification)
Supervised learning: Contains the result (labels) data
Category: Sample tags for discrete variable
regression: sample labels belong to a continuous variable
Classification:
large generative model (probability model) results output probability
discriminant model (non-probabilistic) is determined directly from the feature score
Joint probability distribution
