- 目的,
- 关键特征,用实时竞价的方式链接广告和上下文,用户
- 按照展示收取费用
- cookie mapping 三个核心问题,谁发起?在哪发起?谁存mapping表?
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Plugin architecture based on "sub-class discovery method" described in a talk | |
by Dr. André Roberge. The talk is called "Plugins and monkeypatching: | |
increasing flexibility, dealing with inflexibility". | |
Here's the talk recorded at PyCon 2009 in Chicago: | |
http://us.pycon.org/2009/conference/schedule/event/47/ | |
Plugin writer HOWTO: | |
1. subclass "Base" | |
2. implement sayHi() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import sys | |
import traceback | |
import json | |
# Path for spark source folder | |
os.environ['SPARK_HOME']="/Users/jilu/Downloads/spark-1.3.0-bin-hadoop2.4" | |
# Append pyspark to Python Path |
-
长尾的(a,u,c)组合
-
Multi-arm Bandit问题
有限个arms,代表每一个a,每一个有有限的确定收益
目的是在每个时刻t,我们必须从arms中选择一个,最终目标是优化整体收益
- 挑战 海量空间需要被探索,每个arm期望收益是动态变化的
-
L-BFGS,Quasi-Newton方法的一种,Hession阵要正定 将n x n的矩阵用另外一种方法近似
-
ADMM,减少迭代,拉格朗日的解法(augmented lagrangian)
- p(click | a,u,c)
- 回归(Regression)比排序(Ranking)更合适
- cold-start,利用广告层级结构(creative,solution,campaign,advertiser),以及广告标签对新广告点击率做预测
- 捕获点击率的动态特性,快速调整特征,快速调整模型
视角1,逻辑回归是广义线性模型在Binomial error情形的特例
视角2,最大熵模型在类数目等于2的情形特例
paxos算法
可以视为query为a,对(u,c)进行检索,由于(u,c)联合空间过大,可以单独对u,c检索
NewerOlder