NOTE 2

1. Describing objects by their Attributes.

Many opinion classifiers are created by adopting
a "kitchen sink" approach that throws together
a variety of features. But in many cases adding
new types of features does not improve performance.

kitchen sink approach 指把所有特征放在一起，就像把所有买来的菜放在厨房的水槽里。

We use the kitchen sink approach to feature extraction – we extract all features that we think might be important. But learning algorithms often work better on smaller feature sets. Large feature sets can lead to overfitting the data, or unimportant features can flat-out confuse learning algorithms.

So we use feature selection to find the most informative features for us.

We can use MIS to tell us how informative a given feature is, or how well it can predict the unroll factor.

所以 feature selection 是非常重的。

graphical model 能不能用于 feature selection?

2.  How to represent shapes and locations? One simple way is to constuct local histograms and concatenate these together. Is there other (efficient) shape features?

logistic regression: 收藏的程序