1.1 Introduction
- machine learning
model == learning algorithm
The term substantially 1.2
- A collection of records: data set (data set)
- Example (instance) / sample (sample): a set of records
- Property (attribute) / feature (feature)
- Attribute value (attribute value)
- Attribute space (attribute space) space / sample space (sample space) / Input: attribute space spanned
- Feature vector (feature vector): the attribute a1 / a2 / a3 as three coordinate axes, which span a three-dimensional space is described for example, each example can be found in the own coordinate space. Each coordinate point corresponds to a vector space, it is referred to the example of feature vector.
- 维数(dimensionality):
- D = {x1, x2 ... xm} denotes the m-th instance of the data set
- xi = (xi1, xi2 ... xid ) a d-dimensional vector in the sample space X, xi belongs to X, xij is the value of the property, d is the dimension of the sample xi. How much property is the number of dimensions .
- Learning (learning), training (training): data obtained from the model school process.
- Hypothesis (hypothesis): corresponding to the model learn some underlying data about the law.
- The truth, the real (ground-truth): Potential law itself, the learning process is to find out the truth or approximation .
- Model, the learner (learner)
- Forecast (prediction)
- Mark (label)
- The sample (example): has the label of instance / sample. With (xi, yi) denotes the i th sample.
- , Mark space output space (label space): the label set.
- Classification (classification): prediction of discrete values, such as "melon good" and "bad melon"
- Regression (regression): the predicted continuous value, e.g. 0.95,0.37 watermelon maturity.
- Test (testing): After the learned model, using its process of prediction.
- Test samples (testing sample): the prediction of the sample. For example, the learn F, test sample x, which can be obtained prediction flag y = f (x).
- Clustering (clustering): watermelon in the training set into groups, each group of one cluster (cluster). ** cluster automatically form may correspond to some of the underlying concepts in the "light-colored melon", "
1.3 hypothesis space
- Space version (version space)
- Induction (induction): from the particular to the general "generalization" (generalization) process, from specific facts summed up the general rule.
- Deduction (deduction): from the general to the specific "specialization" (specialization) process, from basic principles to deduce the specific circumstances. For example, according to a mathematical axiom deduction.
- Inductive learning
- Narrow inductive learning (training data obtained from the concept, also known as concept learning)
- Generalized inductive learning (learning from the example of)
The most basic concept learning: Boolean concept learning, 0/1 Boolean value.
1.4 induction preferences
- why