Linda_ak-CSDN博客

原创深度学习的自然语言处理-课程笔记-1

课程链接:https://www.bilibili.com/video/av15892671/?spm_id_from=333.788.b_636f6d6d656e74.29（b站） https://www.youtube.com/watch?v=OQQ-W_63UgQ&list=PL3FW7Lu3i5Jsnh1rnUwq_TcylNr7EkRe6&index=2&t=0...

2019-01-10 22:29:45 210

原创 fasttext小试牛刀

解读源码：原始输入其实是one-hot，wi其实就是lookuptable，所有的wordvector加和求平均直接得到隐藏层，隐藏层经过wo矩阵得到输出，再进行softmax。需求一：已经有其他样本训练好的词向量，想直接用，迁移学习如果有预训练好的wordvec，利用参数-pretrainedVectors 指定文件就行，这个是用与训练的文件初始化参数，迭代过程中词向量仍会变。如果...

2019-07-10 20:54:03 184

原创深度学习的自然语言处理-课程笔记-2-word2vec

照例附上大牛的笔记：http://www.hankcs.com/nlp/word-vector-representations-word2vec.html 这节课主要是讲word2vec。两种常见的算法是： (1)skip-grams(SG) predict context words given taget(position independent)，预测上下文 (2)conti...

2019-01-13 21:11:25 163

原创神经网络学习笔记

1、书籍《python神经网络编程》这本书只认真读完了第一章，大概80页内容，就是简单介绍了下三层前馈神经网络，以及推导了利用梯度下降法来反向传播误差，更新权重，即误逆差传播法（error backpropagation, BP)。有点基础的，这本书还是不推荐买了，完全可以直接阅读《深度学习》(周志华)第5章神经网络的内容。 2、上手练习 2.1 利用tensorflow进行文本分类参...

2018-11-28 20:51:17 215

原创机器学习常见面试题

校招季整理，会一直更新 1、判别式模型和生成式模型判别方法：由数据直接学习决策函数 Y = f（X），或者由条件分布概率 P（Y|X）作为预测模型，即判别模型。生成方法：由数据学习联合概率密度分布函数 P（X,Y）,然后求出条件概率分布P(Y|X)作为预测的模型，即生成模型。由生成模型可以得到判别模型，但由判别模型得不到生成模型。常见的判别模型有：K近邻、SV

2016-07-22 10:26:17 1540

原创常见算法与数据结构整理

校招季自己整理的相关知识点，会一直更新。一、动态规划主要思路：先确定优化方程，再确定递推关系 1、最长公共字串（LCS) X，Y的公共字串为Z，Z中的字符串在X，Y中不一定要连续出现，但是是X，Y中递增出现。子问题：C[i，j]表示Xi和Yj的最长公共子串，则若Xi=Yj，C[i，j]=c[i-1，j-1]+1，否则C[I,j]=max(C[i-1.j],C[i

2016-07-21 16:45:37 2433

原创 Hadoop自定义linereader，实现按行分块

Hadoop自定义linereader，实现按行分块，map方法按行处理

2015-12-23 22:22:15 1659 1

数字图像处理

数字图像处理，冈萨雷斯的版本，Matlab版本，适合详细读。。。

2015-06-22

马尔可夫状态分类

马尔可夫模型的入门知识，介绍的很详细，我入门就是看的这个。。。

2015-06-22

Given an increasing sequence S of N integers, the median is the number at the middle position. For example, the median of S1={11, 12, 13, 14} is 12, and the median of S2={9, 10, 15, 16, 17} is 15. The median of two sequences is defined to be the median of the non-decreasing sequence which contains all the elements of both sequences. For example, the median of S1 and S2 is 13. Given two increasing sequences of integers, you are asked to find their median.

2013-09-19

题目1003：A+B

给定两个整数A和B，其表示形式是：从个位开始，每三位数用逗号","隔开。现在请计算A+B的结果，并以正常形式输出。

2013-09-19

题目1002：Grading

Grading hundreds of thousands of Graduate Entrance Exams is a hard work. It is even harder to design a process to make the results as fair as possible. One way is to assign each exam problem to 3 independent experts. If they do not agree to each other, a judge is invited to make the final decision. Now you are asked to write a program to help this process. For each problem, there is a full-mark P and a tolerance T(<P) given. The grading rules are: • A problem will first be assigned to 2 experts, to obtain G1 and G2. If the difference is within the tolerance, that is, if |G1 - G2| ≤ T, this problem's grade will be the average of G1 and G2. • If the difference exceeds T, the 3rd expert will give G3. • If G3 is within the tolerance with either G1 or G2, but NOT both, then this problem's grade will be the average of G3 and the closest grade. • If G3 is within the tolerance with both G1 and G2, then this problem's grade will be the maximum of the three grades. • If G3 is within the tolerance with neither G1 nor G2, a judge will give the final grade GJ.

2013-09-19