Covariance measures the average product of centered deviations and carries the units of both variables. Dividing by their positive standard deviations gives dimensionless correlation. Positive rescaling changes covariance but preserves correlation; zero correlation alone does not imply independence.协方差是两个变量相对各自均值的偏差之积的平均,带有两者的单位。除以各自正的标准差,就得到无量纲的相关系数。正比例缩放会改变协方差,却不改变相关系数;相关系数为零也不一定意味着独立。
The left cloud stretches horizontally as changes. The right cloud subtracts each mean and divides by its standard deviation, so it keeps the same shape and . These equally weighted constructed points illustrate normalization; the moving gold markers connect corresponding positions.左侧点集随 改变而横向伸缩;右侧减去各自均值、再除以标准差,形状和 因而保持不变。这里使用等权重的构造点集演示标准化,移动金点标出两边的对应位置。