Identifying implicit, previously unknown, potentially useful patterns in data.
- Interactively or automatically done
- Descriptive or predictive
找到不明的，之前未知的，且潜在有价值的数据模式。
- 交互式地或自动地完成
- 描述性或预测性
Programs that induce structural descriptions from observations.
- Supervised learning: Based on labeled examples & used for predicting labels of new observations.
- Unsupervised learning: Based on unlabeled data.
通过程序从各观察中得到结构性描述。
- 监督学习：基于标记了的例子并用来为新观察预测其标记。
- 非监督学习：基于未标记的数据。
if-then Rules
- Classification Rule: Predicts value of a given attribute
- Association Rule: Predicts value of arbitrary attribute
if-then规则
- 分类规则：预测已知的属性
- 关联规则：预测任意的属性
Structures that can be learned. 可用于学习的结构。
Individual, independent examples of a concept, possibly corrupted by noise.
单独的，独立的概念例子，可能被噪音损坏。
Measuring aspects of an instance. 实例中可供测量的方面。
Output of learning scheme. 学习方案的输出。
Supervised, outcome is called class, success is subjectively measured.
监督学习，输出被称为类，成功的预测可被客观地测量。
Unsupervised, no class is specified, any interesting kinds of structure.
非监督学习，没有明确的类，任何有趣的结构都可以。
Unsupervised, find similar groups of items, success is subjectively measured.
非监督学习，寻找能够相似的样本成组，成功可以被客观地测量。
Supervised, regression, predict a numeric value rather than a discrete class
监督学习，回归，预测一个数值量而不是离散类
Process of flattening, any finite set of finite relations.
对一些存在有限关系的有限集合的一个平面化的处理过程。
- Nominal: Values are distinct symbol, no relation is implied, only equality test
- Ordinal: Impose order on values, no distance between data
- Interval: Numeric, fixed & equal unit, difference is accepted
- Ratio: Defined non-arbitrary zero point, multiplication & division are permitted
- 名目：值为独特的符号，之间没有隐晦任何关系，只能平等地测试
- 序数：值间有顺序，但没有距离
- 区间：数值类型，固定和相等的单位，可以取差值
- 比率：定义了一个并不随便的零点，可以进行乘除
@relation relation-name
@attribute attribute-name-1 {nominal/ordinal}
@attribute attribute-name-2 numeric/string/date
@data
attribute-name-1, attribute-name-2
-
Sparse Data: In some applications most attribute values in a data set are zero.
0, x, 0, 0, 0, 0, y 0, 0, 0, w, 0, 0, 0ARFF supports Sparse Data as:
{1, x, 6, y} {3, w}
- Indicated using
?for unknown, unrecorded, irrelevant value. - Missing value may have significance in itself.
- 使用
?来表示未知的，未记录的，不相关的值。 - 缺失值有时有重要意义。
One class is much more prevalent than the others.
**Solution: ** Assign different costs to different types of misclassification
一个类的数据明显比其他类更加流行。
解决方案： 为未分类的不同类型赋予不同的权值。
The simplest way of representing the output of learning. 最简单的表达学习输出的方式。
Output is a sum of all attributes with its weights. 输出是各加权重属性求和。
Divide-and-conquer approach to processing the training data produces a decision tree.
- Internal nodes in the tree test attributes
- Attribute values are compared to constant
- Comparing values of two attributes
- Using a function of one or more attributes
- Leaves assign classification or probability distribution to instances
- To make prediction, instance with unknown class label is routed down the tree
使用分治法来训练数据以建立一个决策树。
- 内部的节点用于树测试属性
- 属性值与常数比较？
- 比较两个属性值
- 使用函数的一个或多个属性？
- 为叶节点赋予分类或实例的概率分布
- 预测时，有未知类的实例由根节点到叶节点决策
- Nominal:
- One branch for every value, number of children is equal to number of attributes.
- Just test once only
- (Alternative) Division into two subsets of nominal attribute values
- Numeric:
- Test whether value is greater or less than constant
- Test more than once
- (Alternative) Three-way split
- 名目：
- 一个分支对应所有值，子节点的数目与属性数目一致。
- 只测试一遍。
- 另外的方法：分裂到两个名目子集
- 数值：
- 测试值是否大于或低于一个常量
- 测试多次
- 另外的方法：分裂成三个分支
Absence of value have some significance?
- Yes! Missing should be a separate value
- No! Missing must be treated in a special way
- Solution A: Assign instance to most popular branch
- Solution B: Split instance into pieces with one piece per branch extending from node
缺失值很重要吗？
- 是的！缺失值需要是分离的值。
- 不是！缺失值必须通过一个特殊的方法对待
- 方法A：为最流行的分支分配缺失值的实例
- 方法B：分裂实例，从节点一个分支衍生一部分缺失值的实例出来
Popular alternative to decision trees.
Antecedent (Pre-condition): A series of tests, logically ANDed together.
Consequent (Conclusion): Class or probability distribution assigned by rule, logically ORed together.
另一种流行
先例 (先决条件)：一系列的测试，通常以和逻辑合并
后果 (结论)：类，或者概率分布，通常以或逻辑合并
One rule for each leaf.
- Antecedent contains a condition for every node on the path from root to leaf.
- Consequent is class assigned by the leaf.
Produces rules that are unambiguous. Rules can be unnecessarily complex which need pruning.
一个叶节点，一条规则。
- 先例包含了从根到叶的所有节点的条件。
- 结论是叶节点的类。
生成的规则并不模糊，规则没有必要那么复杂，所以需要修剪。
- More difficult than from trees to rules.
- Symmetry needs to be broken.
- Corresponding tree contains identical subtrees.
(Replicated Subtree Problem)
- 比起从树到规则更难
- 对称需要被打破
- 相应的树包含了相同的子树
重复子树问题
Question: Are rules independent pieces of knowledge?
Two ways of executing a rule set:
- Ordered set of rules (Decision list): Order is important for interpretation.
- Unordered set of rules: Rules may overlap and lead to different conclusions for the same instance.
问题：规则是独立的知识片段吗？
两种执行规则集的方式：
- 顺序的规则集：顺序对于解释极为重要。
- 无序的规则集：规则可能覆盖或者导致对于相同的实例产生不同的结论。
- Predict any attribute and combinations of attributes.
- Not intended to be used together as a set.
Problem: Immense number of possible associations.
Solution: Restrict to show only the most predictive associations with high support and high confidence.
- 预测任意属性或者属性集合
- 并不意味着被用作一个集合
问题：可能存在大量的可行关联规则。
解决方案： 限制只输出高支持度和高置信度的预测性关联规则。
Support: Number of instances predicted correctly.
Confidence: Number of correct predictions, as proportion of all instances that rule applies to.
Normally minimum support and confidence are pre-specified.
支持度：正确预测的实例个数。
置信度：正确预测的实例的个数，占据总实例的比例。
通常最小支持度和置信度都是预先指定了的。
Lazy learning, AKA Rote Learning
- Training instance are searched for instances which are most closely resemble test instance.
- The instances themselves represent the knowledge.
Methods: Nearest-Neighbor, K-Nearest-Neighbor
懒惰学习，又称死记硬背法学习
- 训练的实例由那些最接近于测试实例的实例构成。
- 实例本身就表达了知识。
方法：Nearest-Neighbor, K-Nearest-Neighbor
- Numeric: Euclidean distance with normalization
- Nominal: 1 - different, 0 - equal
- 数值：规范的欧氏距离
- 名目：1 - 不同， 0 - 相同