机器学习方面的英文电子书快速过一遍，看看有没有适合于内容推荐引擎的新想法~~~

Machine Learning with Spark-Packt Publishing(2014) epub
 Feature hashing
Feature hashing is a technique to deal with high-dimensional
data and is often used with text and categorical datasets where
the features can take on many unique values (often many
millions of values). In the previous chapters, we often used the
1-of-K encoding approach for categorical features, including text.
While this approach is simple and effective, it can break down
in the face of extremely high-dimensional data.


  2016/2/19 16:34:21
Feature hashing works by assigning the vector index for a
feature based on the value obtained by hashing this feature to
a number (usually, an integer value) using a hash function. For
example, let's say the hash value of a categorical feature for
the geolocation of United States is 342
. We will use the
hashed value as the vector index, and the value at this index
will be 1.0 to indicate the presence of the United States
feature. The hash function used must be consistent (that is, for
a given input, it returns the same output each time).


  2016/2/19 16:34:50
高维的特征首先是一个很长的01比特串？然后再进一步hash为一个小一点的整数？


  2016/2/19 16:35:19
http://pan.baidu.com/s/1kUrXpHl
16:36:42
  2016/2/19 16:36:42
这里似乎原始的特征向量仍然是使用的整个字典作为每一个维度？？？


  2016/2/19 16:37:57
缺点：（1）无法区分那个维更重要一点；（2）hash方法都存在冲突问题——不过后者一般可以用bloom filter来弥补吧


  2016/2/19 16:38:36
A key paper that introduced the use of hashing for feature
extraction and machine learning is:
Kilian Weinberger, Anirban Dasgupta, John Langford
, Alex
Smola, and Josh Attenberg. Feature Hashing for Large
Scale Multitask Learning. Proc. ICML 2009, which is
available at
http://alex.smola.org/papers/2009/Weinbergeretal09.pdf
16:40:12
  2016/2/19 16:40:12
https://www.gracecode.com/posts/3148.html


  2016/2/19 16:40:22
我记得NFC一开始是nokia搞出来的吧？？？


  2016/2/19 16:41:38
http://bohaishibei.com/post/16268/
16:45:57
  2016/2/19 16:45:57
Word2Vec models: 把每个单词表示为一个向量？？？这有点LSI的感觉


  2016/2/19 16:46:24
One of the main academic papers underlying Word2Vec is
Tomas Mikolov
, Kai Chen, Greg Corrado, and Jeffrey Dean
. Efficient Estimation of Word Representations in Vector
Space. In Proceedings of Workshop at ICLR, 2013.
It is available at http://arxiv.org/pdf/1301.3781.pdf.


  2016/2/19 16:47:50
唉，看了半天，也没感觉有多少新意
16:52:11
  2016/2/19 16:52:11
Machine Learning with R - Second Edition-Packt Publishing (2015).pdf: 评估模型性能：Beyond accuracy – other measures of performance


  2016/2/19 16:52:49
The kappa statistic
The kappa statistic (labeled  Kappa in the previous output) adjusts accuracy by
accounting for the possibility of a correct prediction by chance alone. 


  2016/2/19 16:53:20
Sensitivity and specificity（？）


  2016/2/19 16:53:36
Precision and recall


  2016/2/19 16:53:55
The F-measure
A measure of model performance that combines precision and recall into a single
number is known as the F-measure
16:54:42
  2016/2/19 16:54:42
提高性能（集成学习）：bagging、boosting、随机森林


  2016/2/19 16:54:53
好吧，让我再看看另外一本书


  2016/2/19 16:56:10
Machine Learning An Algorithmic Perspective (2nd Edition)-CRC Press 2015 这本书最后居然讲到到DBN，偏学术一点，似乎没有针对对文本挖掘的例子？
16:58:08
  2016/2/19 16:58:08
Machine Learning - A Probabilistic Perspective 这本书就更偏学术一点，Monte Carlo推理？还有什么MCMC


  2016/2/19 16:58:46
Particle filtering (PF) is a Monte Carlo, or simulation based, algorithm for recursive Bayesian
inference. That is, it approximates the predict-update cycle described in Section 18.3.1. It is
very widely used in many areas, including tracking, time-series forecasting, online parameter
learning, etc. 


  2016/2/19 16:59:51
靠，这本书最后一章讲到了DL，还挺详细的，MIT的这本2012年的书。。。。
17:00:11
  2016/2/19 17:00:11
换一本，Machine Learning in Python_ Essential Techniques for Predictive Analysis-Wiley (2015)


  2016/2/19 17:01:25
Penalized Linear Regression（解决OLS过分适配的问题）
17:03:53
  2016/2/19 17:03:53
PLR是加罚分函数的意思吗？


  2016/2/19 17:03:57
换一本，Machine Learning_ Hands-On for Developers and Technical Professionals-Wiley (2015)


  2016/2/19 17:04:25
Spring XD？


  2016/2/19 17:05:19
虽然机器学习语言用到R，但是好像我听说R并不适合分布式大规模可扩展型的任务


  2016/2/19 17:05:51
Spark呢，一则Scala学习曲线高了点，二来部署代价也高（需要大内存的机器）
17:06:43
  2016/2/19 17:06:43
这本书感觉一般般，似乎阅读难度不大


  2016/2/19 17:06:56
再换一本，Machine Learning_A Bayesian and Optimization Perspective-Academic Press(2015)
17:08:59
  2016/2/19 17:08:59
什么是SPARSITY-AWARE LEARNING？感觉这里面tricks很多。。。
17:12:04
  2016/2/19 17:12:04
MC采样似乎可以用在A/B测试的分类上，但是这本书的代码例子居然是用Matlab写的，基本没有涉及文本挖掘（主要是关于图像识别的）


  2016/2/19 17:12:20
Machine Learning_The Art and Science of Algorithms that Make Sense of Data-Cambridge University Press(2012)


  2016/2/19 17:12:26
看看这一本怎么样


  2016/2/19 17:12:54
噢，这一本不行，太low了，最近刚翻译，我有买的
17:16:03
  2016/2/19 17:16:03
Python Machine Learning-Packt 2015 ： 维度压缩（LDA和PCA），bag-of-words（其实就是TF-IDF全局方法，没啥新意）
17:19:07
  2016/2/19 17:19:07
Learning Data Mining with Python-Packt Publishing 2015：DNN2个包（Theano、Lasagne）；文本聚类，k-means + Online learning??


  2016/2/19 17:20:25
Efficient Learning Machines_Theories, Concepts, and Applications for Engineers and System Designers-Apress(2015): Among these
models, cortical algorithms (CAs) have emerged as a
biologically inspired approach, modeled after the human visual
cortex, which stores sequences of patterns in an invariant form
and recalls those patterns autoassociatively.


  2016/2/19 17:20:34
CA？有点意思嘛
17:21:09
  2016/2/19 17:21:09
CA感觉是DNN的一种？


  2016/2/19 17:22:42
Packt.Building Machine Learning Systems with Python.2Ed.2015 ： Topic Model，LDA