韭菜地

转载 Edit Distance

The minimum edit distance between two strings is the minimum number of editing operations Insertion Deletion Substitution needed to transform one into other.Defining Min Edit DistanceFor two Strings:X

转载 word2vec

Main idea of word2vecInstead of capturing cooccurrence counts directlyPredict surrounding words of every wordFaster and can easily incorporate a new sentence/document or add a word to the vocabulary

转载 sense2vec - a fast and accurate method for word sense disambiguation in neural word embeddings.

Despite these advancements, most word embedding techniques share a common problem in that each word must encode all of its potential meanings into a single vector.This technique is inspired by the work

转载 End-To-End Memory Networks

Single LayerA layer has two memroy: input memory,output memory. Parameters are A∈Rd×|V|A \in \Bbb R^{d \times |V|},B∈Rd×|V|B \in \Bbb R^{d \times |V|},C∈Rd×|V|C \in \Bbb R^{d \times |V|},W∈R|V|×dW \in

转载 the equations of backpropagation

backpropagation

转载 Rectified Linear Unit (ReLU)

The Rectified Linear Unit (ReLU) computes the function f(x)=max(0,x)f(x)=max(0,x), which is simply thresholded at zero.There are several pros and cons to using the ReLUs:(Pros) Compared to sigmoid/tan

转载 正则表达式

表1.常用的元字符代码说明.匹配除换行符以外的任意字符\w匹配字母或数字或下划线或汉字\s匹配任意的空白符，包括空格，制表符(Tab)，换行符，中文全角空格等\d匹配数字\b匹配单词的开始或结束^匹配字符串的开始$匹配字符串的结束

转载 What are the continuous bag of words(CBOW) and skip-gram?

Both architectures describe how the neural network “learns” the underlying word representations for each word. Since learning word representations is essentially unsupervised, you need some way to “cre

转载 求导公式

转载 second derivative & Hessian matrix

We are also sometimes interested in a derivative of a derivative. This is known as a second derivative. For example, ∂2∂xi∂xjf\frac{\partial ^2}{\partial x_i \partial x_j}f is the derivative with respe

转载 Determinant

The determinant of a square matrix, denoted det(A)det(A), is a function mapping matrices to real scalars. The determinant is equal to the product of all the matrix’s eigenvalues. The absolute value of

转载 linear regression example

------------------------------------------------------------------------ example-linear-regression.lua-- -- This script provides a very simple step-by-step example of-- linear regression, using Tor

转载 Improving the way neural networks learn

LINKWhy sigmoid + quadratic cost function learning slow?The quadratic cost function is given by C=(y−a)22(1)C=\frac {(y-a)^2}{2} \tag 1 where aa is the neuron’s output. a=σ(z)a=\sigma (z), where z

转载 Brief overview of backward and forward

Let’s say we only feed in one data point.out = model:forward(xix_i) computes fw(xi)f_w(x_i) where fwf_w is our model with its current parameters ww, and stores the result in out.loss = criterion:forw

转载 Torch7 Tensor slicing

------------------------------------------------------------------------ slicing.lua-- -- This script demonstrates tensor slicing / manipulation.-- To run this script, simply do:-- $ th A_slicing.l

转载 Understanding the difficulty of training deep feedforward neural networks

softsignA newly proposed activation function (Bergstra et al., 2009) called the softsignsoftsign: x/(1+|x|)x/(1+|x|) The softsignsoftsign is similar to the hyperbolic tangent but its tails are quadra

转载 Gated Recurrent Units (GRU)

Illustration: update gate zt=σ(Wzxt+Uzht−1) z_t=\sigma(W^{z}x_t+U^{z}h_{t-1}) Update gate zz controls how much of past state should matter now. If zz close to 1, then we can copy information in

原创 Torch7 doc

nn.SplitTable() -- (N)dim Tensor -> table of (N-1)dim Tensorsnn.JoinTable() -- table of (N-1)dim Tensors -> (N)dim Tensor

转载 LSTM

The following graph shows such a lstm memory block. The blue arrows are the peephole connections. So the gates “see” the cell state(s) even if the output gate is closed.In the following a memory bl

转载 Torch7 Serialization

Torch provides 4 high-level methods of serialize/deserialize arbitrary Lua/Torch objects. These functions are just abstractions over the File object, and were created for convenience (these are very co

转载 pcall

If you need to handle errors in Lua, you must use the pcall (protected call) function to encapsulate your code.The pcall function calls its first argument in protected mode, so that it catches any erro

转载 Defining your own Neural Net Module

Modules are bricks to build neural networks. A Module is a neural network by itself, but it can be combined with other networks using container classes to create complex neural networks. Module is an a

翻译 The "ReQU" unit

Here, we’ll impolement a made-up activation function that we’ll call the Rectified Quadratic Unit(ReQU). Like the sigmoid and ReLU and several others, it is applied element-wise to all its inputs:zi=I[

转载 Jacobian

来源 Given a set y=f(x)\mathbf{y=f(x)} of nn variables x1,...,xnx_1,...,x_n, written explicitly as y=⎡⎣⎢⎢⎢⎢⎢f1(x)f2(x)⋮fn(x)⎤⎦⎥⎥⎥⎥⎥\mathbf y = \left[\begin{array}{c}f_1(\mathbf x)\\f_2(\mathbf x)\\

转载 optim package

Optimization packageThis package contains several optimization routines for Torch. Each optimization algorithm is based on the same interface:x*, {f}, ... = optim.method(func, x, state)where:func: a

转载 Energy Function

能量函数

原创 Analysis of Discrete Data|STAT 504

来源Matrix Algebra ReviewTo multiply two vectors with the same length together is to take the dot product, also called inner product. Advanced Matrix Properties

翻译 matplotlib趋势线

matplotlib画散点图十分的方便，使用numpy的polyfit函数实现Excel中的趋势线功能也是很简单的。API参考地址# plot the data itselfpylab.plot(x,y,‘o’)# calc the trendline (it is simply a linear fitting)z = numpy.polyfit(x, y, 1)p = numpy.pol

转载 matplotlib嵌入wxpython panel

matplotlib嵌入wxpython panel

转载 php 中关于 fopen 如何打开或创建中文文件的使用说明

在IT编程界，中文乱码一直是个头疼的问题。但是，只要时常总结出现的问题，以后再遇到相同的情况，直接在博客里查找就可以了。今天，有一个朋友在我的另一篇关于PHP作下载功能的博文(www.gretheer.com/2013/06/php-download.html)里发现，在浏览器地址栏(URL)中输入中文，下载功能就出现了异常。经过调试代码，我发现是 fopen 打开中文文件名的文件引起

翻译 机器学习算法的分类

Below are 5 classes of machine learning algorithm that can be used to group algorithms by structure and learning style and 3 examples of algorithms for each class.1) Regression: linear regressio

翻译 升级到VMWARE11后，vmcore/vmm/main/physMem_monitor.c:1123错误解决

用记事本打开出错虚拟机的vmx文件，添加一行smc.version = 0，即可解决此错误。

转载 深度学习算法的几个难点

1、局部最优问题。深度学习算法的目标函数，几乎全都是非凸的。而目前寻找最优解的方法，都是基于梯度下降的。稍微有点背景知识的人都知道，梯度下降方法是解决不了非凸问题的。因此，如果找到最优解，将是深度学习领域，非常值得研究的课题。andrew在google的工作，也就是那只猫，其实训练过程是让人很费解的。为了缩短训练时间，项目组采用了分布式训练的方式。采用了1000台计算机，在不同的计

转载 SEO都做什么?

SEO入门人士都以为SEO是一个多么神圣的行业，里面有多么高深的技巧，不知道SEO究竟做什么，本文章将对SEO做什么？做以简单的介绍．以便轻轻的抹去大家对SEO的神密面纱．  下是SEO搜索引擎优化的大体内容： 1 代码优化：网站代码简洁，加载速度更快，节约资源大大提高公司在互联网上的形象。  2 图片优化：图片、文字以及LOGO都有明确的标注，潜在客户在搜索图片或视频时，我们的网站出

转载 JSP中文及传中文参数乱码解决方法小结

在使用JSP的过程中，最使人头疼的一个问题就是中文乱码问题，以下是我在软件开发中遇到的乱码问题以及解决方法。 　　1、JSP页面乱码　　这种乱码的原因是应为没有在页面里指定使用的字符集编码，解决方法：只要在页面开始地方用下面代码指定字符集编码即可，　　2、数据库乱码　　这种乱码会使你插入数据库的中文变成乱码，或者读出显示时也是乱码，解决方法如下： 在数据库连接字符串中加入编码字符集

转载 java 正则表达式 选择多行内容

Html代码titletitle2 用正则表达式选取...中内容，表达式如下：Java代码[/s/S]*? 转自：http://stupid.javaeye.com/blog/153074

转载 解决 若要使用 Windows Media Player，请从 Windows 上注销，重新登录，然后启动 Windows Media Player。

WIN2K3 WMP11 好几次都碰到这问题了，每次都是重新装一下，今天又来了终于把我给惹急了，捕获一下注册表，最后居然是因为一个键值Windows Registry Editor Version 5.00[HKEY_LOCAL_MACHINE/Software/Microsoft/Active Setup/Installed Components/{6BF52A52-394A-11d

转载 PowerPoint字体“随身带”

      我们经常会遇到这样的情况：由于每台计算机中安装的字体文件不同，在一台计算机上制作好的PowerPoint课件在另一台计算机上打开时，设定的字体会发生改变，影响了课件的播放效果。  　　其实，如果大家使用的是PowerPoint 2002/2003，只要在保存时“嵌入TrueType字体”，就可以让PowerPoint将字体“随身带”了。设置方法如下。  　　打开一个PowerP

原创 特征选择简介

      特征选择是从一组特征中去掉冗余或不相关的特征来进行降维。它可以从原始特征中找到最优特征 ，这些被选择出的特征保留了数据集的主要信息，为分析高维的特征问题提供了便利， 避免或减少原始特征中不相关的信息所带来的识别过程复杂化。 特征选择包含了两个方面， 一方面是试图从特征集合中选择最为有效的特征子集， 另一方面， 选择合适的评价准则确定所选特征的有效性。 一般地， 由于特征之间存在一定的冗