简介
Overfitting & Underfitting¶
-
Overfitting Model is too complex to be generalized to new data
-
Underfitting Model is to simplistic to accurately capture the patterns of the training set, thus fail to generalized to new data
Unsupervised Learning¶
Dimensionality Reduction¶
-
create new repr of data that are easily understood by computer algo than the original repr of the data
-
is a frequent application of transformation, which identify new ways to represent data and generalizes important properties with fewer features
Algorithm¶
-
PCA(Principal ...)
Identify the hyperplane（超平面） closest to data distribution and project the data onto the plane
Clustering¶
Algorithm¶
-
K-Means Clustering
For a give dataset, find k cluster centers based on distances between
Reinforcement learning¶
RL is learning what to do, e.g. how to map the current situation into actions to maximize the gain
Q-Learning¶
Transition function of Q-Learning could be described as:
