机器学习 5 machine learning advice

1. Bayesian regularization

2. online learning

stochastic gradient descent: 随机梯度下降

3. ML advice

a. more training examples => fix high variance
b. Trying a smaller set of features => fixes high variance.
c. Trying a larger set of features  => fix high bias.
d. adding email features => fix high bias.
e. run gradient descent for more iterations  => fixes the optimization algorithm

f. try Newton's method => fixes the optimization algorithm
g. using a different value for lambda =>  fixes the optimization objective
h. changing to an SVM is also another way of trying to fix the optimization objective

see more in  http://download.csdn.net/detail/nomad2/3759561