Classification
Classification answers yes or no.
Logistic regression
A line through a sigmoid gives a probability.
The points and the model's probability
- yes (positive)
- no (negative)
- decision line
- wrong on validation
Evaluation: the threshold decides
Lower the threshold to find more, at the price of more false alarms.
Support vector machines
The widest street between two classes, held up by the support vectors.
The street (click to add a point)
- yes
- no
- decision line
- the street's edges
- support vector
- on the wrong side
- Support vectors
- –
- Training errors
- –
- Street width
- –
A new point outside the street changes nothing.
The RBF kernel lets the same algorithm draw curves.
More than two classes: OvR, OvO and softmax
Three ways to go from two classes to many.
The plane split into classes
Lines are the binary models' boundaries.
Confusion matrix, validation points
The point you clicked