Machine Learning with Swift

Logistic regression

You can come up with a lot of problems a linear classifier has. One of them is that many datasets just cannot be separated properly by a straight line:

Figure 7.2: Linear separable (on the left) and nonlinear separable (on the right) data. The decision boundary as shown in the dashed line. Source: Mykola Sosnovshchenko.

Another problem is that a linear regression line can predict negative values or values greater than one for some samples even though we know for sure y should be either zero or one. To fix it, we need some function that takes values from [-∞, +∞] and output values in the range from zero to one. One such ...