label

 

an image is represented as one large 3-dimensional array of numbers

total numbers : 248*400*3 (wide*height*rgb(3 channels))

each number ranges from 0~255

 

Target: be invariant to the cross product of all these variations, while simultaneously retaining sensitivity to the inter-class variations.

take an array of pixels that represents a single image and assign a label to it

pipeline: input, train, evaluation

 

L1 distance  

 

 

L2 distance  

 

Evaluate on the test set only a single time, at the very end. (very carefully)

 

KNN k-nearest neighbor

instead of finding the single closest image in the training set, we will find the top k closest images, and have them vote on the label of the test image. In particular, when k = 1, we recover the Nearest Neighbor classifier. Intuitively, higher values of k have a smoothing effect that makes the classifier more resistant to outliers: