<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[自然语言处理大菜鸟--HsingWang]]></title><description><![CDATA[自然语言处理 & 机器翻译 & http://hlt.suda.edu.cn/~xwang/]]></description><link>https://blog.youkuaiyun.com/hellonlp</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; hellonlp]]></copyright><item><title><![CDATA[[读书笔记] 流畅的python / Fluent Python]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/87260850</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/87260850</guid><author>hellonlp</author><pubDate>Fri, 15 Feb 2019 10:43:34 +0800</pubDate><description><![CDATA[ 

第3章：字典和集合

泛映射类型  可散列的数据类型 An object is hashable if it has a hash value which never changes during its lifetime (it needs a __hash__() method), and can be compared to other objects (it needs an __e...]]></description><category></category></item><item><title><![CDATA[神经机器翻译WMT14英法基准系统 WMT14 English-French Baseline]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/82663198</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/82663198</guid><author>hellonlp</author><pubDate>Wed, 12 Sep 2018 16:37:50 +0800</pubDate><description><![CDATA[ 

最近（2017年以来）的WMT14 English-French Baseline记录

 

1. GNMT

   https://arxiv.org/pdf/1609.08144.pdf

   语料处理：a shared source and target vocabulary of 32K wordpieces

     For the wordpiece models, we ...]]></description><category></category></item><item><title><![CDATA[incremental training/align for word alignment]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/81904599</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/81904599</guid><author>hellonlp</author><pubDate>Tue, 21 Aug 2018 11:29:59 +0800</pubDate><description><![CDATA[ 

The Berkeley Word Aligner

          https://github.com/mhajiloo/berkeleyaligner

          https://github.com/kmccurdy/inc-align

       A small Clojure application which extends the Berkeley Alig...]]></description><category></category></item><item><title><![CDATA[Adaptation for Neural Machine Translation]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/81082510</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/81082510</guid><author>hellonlp</author><pubDate>Tue, 17 Jul 2018 15:50:44 +0800</pubDate><description><![CDATA[目标：7月把NMT adaptation文章通读下

场景：假设储备了大量新闻领域双语语料，少量科技领域双语语料（或者没有），科技领域单语语料（大量、少量或者没有）的大菜鸟翻译公司，接到一个科技领域的翻译项目。如何使用现有资源去尽可能的把科技领域的翻译做好。

问题：这里的科技领域就是in-domain，新闻领域是out-domain。如何使用数量有限的in-domain parallel cor...]]></description><category></category></item><item><title><![CDATA[T2T Transformer 笔记]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/78753205</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/78753205</guid><author>hellonlp</author><pubDate>Fri, 08 Dec 2017 17:44:33 +0800</pubDate><description><![CDATA[讨论：https://www.jiqizhixin.com/articles/2017-06-28-5https://ricardokleinklein.github.io/2017/11/16/Attention-is-all-you-need.html1. Mutli GPU 和 Single 配置的区别https://github.com/tensorflow/tensor2tensor/i]]></description><category></category></item><item><title><![CDATA[2017/0910 实验结果记录]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/78083656</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/78083656</guid><author>hellonlp</author><pubDate>Mon, 25 Sep 2017 11:52:30 +0800</pubDate><description><![CDATA[实验1. Adam+LN（LN没有对context vector和source side的input做）
20000	BLEU score = 0.2947
21000	BLEU score = 0.3159
22000	BLEU score = 0.3109
23000	BLEU score = 0.3298
24000	BLEU score = 0.3196
25000	BLEU score]]></description><category></category></item><item><title><![CDATA[深度学习中Dropout和Layer Normalization技术的使用]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/78079582</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/78079582</guid><author>hellonlp</author><pubDate>Sun, 24 Sep 2017 20:53:33 +0800</pubDate><description><![CDATA[两者的论文：
Dropout：http://www.jmlr.org/papers/volume15/srivastava14a/srivastava14a.pdf


Layer Normalization:  https://arxiv.org/abs/1607.06450


两者的实现（以nematus为例子）：
https://github.com/EdinburghNL]]></description><category></category></item><item><title><![CDATA[神经机器翻译中的领域自适应问题 Domain adaptation for NMT]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/77717570</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/77717570</guid><author>hellonlp</author><pubDate>Wed, 30 Aug 2017 16:00:12 +0800</pubDate><description><![CDATA[相关论文：
1. Stanford Neural Machine Translation Systems for Spoken Language Domains. 


Minh-Thang Luong and Christopher D. Manning. 

 http://www.statmt.org/OSMOSES/Stanford-IWSLT-15.pdf


2. F]]></description><category></category></item><item><title><![CDATA[深度学习训练中cost突然出现NaN]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/77503354</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/77503354</guid><author>hellonlp</author><pubDate>Wed, 23 Aug 2017 17:26:16 +0800</pubDate><description><![CDATA[问题：在深度学习训练中，之前的cost是正常的，突然在某一个batch训练中出现Nan。
网络搜索的资料：
1. How to avoid that Theano computing gradient going toward NaN https://stackoverflow.com/questions/40405334/how-to-avoid-that-theano-computing-]]></description><category></category></item><item><title><![CDATA[网页正文提取工具Beautiful Soup]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/74556091</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/74556091</guid><author>hellonlp</author><pubDate>Thu, 06 Jul 2017 15:35:12 +0800</pubDate><description><![CDATA[Beautiful Soup是什么？
Beautiful Soup is a
 Python library for pulling data out of HTML and XML files. It works with your favorite parser to provide idiomatic ways of navigating, searching, and modifyin]]></description><category></category></item><item><title><![CDATA[深度学习中Dropout策略]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/72649820</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/72649820</guid><author>hellonlp</author><pubDate>Tue, 23 May 2017 16:41:54 +0800</pubDate><description><![CDATA[在写系统时候遇到一个问题：模仿的版本在测试阶段，把所有权重乘以一个保留概率（做法1）。而GJH他看的Keras里面测试阶段什么都不做（做法2），所以比较好奇，怎么做法不同？


但是随手搜了几个中文博客，都是按做法1来的，训练阶段按保留概率随机生成一个保留矩阵(元素要么0-丢弃，要么1-保留)，然后测试阶段所有权重乘以保留概率。


在回头看TensorFlow的Dropout函数ht]]></description><category></category></item><item><title><![CDATA[组内NLP讨论班阶段总结]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/71189611</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/71189611</guid><author>hellonlp</author><pubDate>Thu, 04 May 2017 22:44:32 +0800</pubDate><description><![CDATA[讨论班进行了三期了，效果比想象的要好很多。但也遇到不少问题：
1. 公式推导和编码实现
    困惑是在讲解模型过程中，不可避免地涉及公式推导和编码实现。讨论班上是否对公式推导进行讲授，讨论班后是否组织大家进行编码实现这两点把握不准。
    理想情况下讨论班应该只是成为一个“引子”，通过大家一起的讨论学习对某一个概念有了大致的了解。在后面学习或者实际应用中碰到该概念或者模型知道它是怎么回事]]></description><category></category></item><item><title><![CDATA[TensorFlow Seq2Seq Model笔记]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/66971931</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/66971931</guid><author>hellonlp</author><pubDate>Mon, 27 Mar 2017 15:11:59 +0800</pubDate><description><![CDATA[1. tf.app.run()的疑惑
    http://stackoverflow.com/questions/33703624/how-does-tf-app-run-work
    tf.app类似python中argparse


2.]]></description><category></category></item><item><title><![CDATA[TensorFlow1.0 运行报错 AttributeError: type object 'NewBase' has no attribute 'is_abstract']]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/66582767</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/66582767</guid><author>hellonlp</author><pubDate>Sun, 26 Mar 2017 23:39:35 +0800</pubDate><description><![CDATA[安装TF后，import tensorflow报错：
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opene]]></description><category></category></item><item><title><![CDATA[爱丁堡大学神经机器翻译系统  nematus 使用笔记]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/64921476</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/64921476</guid><author>hellonlp</author><pubDate>Wed, 22 Mar 2017 16:48:03 +0800</pubDate><description><![CDATA[代码准备：
nematus https://github.com/rsennrich/nematus

subword-nmt  https://github.com/rsennrich/subword-nmt]]></description><category></category></item><item><title><![CDATA[强化学习 Reinforcement Learning 资料]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/64648082</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/64648082</guid><author>hellonlp</author><pubDate>Wed, 22 Mar 2017 00:10:00 +0800</pubDate><description><![CDATA[1. 《机器学习》周志华著   第16章
2. https://www.analyticsvidhya.com/blog/2017/01/introduction-to-reinforcement-learning-implementation/
3. https://www.nervanasys.com/demystifying-deep-reinforcement-learning/]]></description><category></category></item><item><title><![CDATA[组内菜鸟学习班]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/60968166</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/60968166</guid><author>hellonlp</author><pubDate>Thu, 09 Mar 2017 21:44:30 +0800</pubDate><description><![CDATA[今天中午和褚师姐孙老师聊天，聊到了专业基础问题。大家都有同感，专业基础太差了。
      11年至今，先后呆过四个组（硕士期间，计算所实习期间，博士期间和诺亚实习期间）。纵览各方豪杰，深感自己基础还是稍显薄弱。基础知识包括：机器学习/数学，专业知识，编程能力和英语读写。对于研究生做科研，自己的感觉就是前两项严重制约个人的发展。其实，目前看来，自己的基础应付毕业是没有大问题的（运气好，水了论文够]]></description><category></category></item><item><title><![CDATA[谷歌 Dynamic Computation Graphs 动态计算图]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/59754049</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/59754049</guid><author>hellonlp</author><pubDate>Thu, 02 Mar 2017 23:47:26 +0800</pubDate><description><![CDATA[上次调研Syntax+NMT（http://blog.youkuaiyun.com/wangxinginnlp/article/details/56488921）发现Sequence-to-Tree 或者 Tree-To-Sequence目前已经火热起来，但是目前绝大部分的深度学习框架（http://blog.youkuaiyun.com/wangxinginnlp/article/details/58670706）只是]]></description><category></category></item><item><title><![CDATA[资料收集：深度学习框架比较]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/58670706</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/58670706</guid><author>hellonlp</author><pubDate>Tue, 28 Feb 2017 21:44:54 +0800</pubDate><description><![CDATA[1. https://en.wikipedia.org/wiki/Comparison_of_deep_learning_software


2. https://deeplearning4j.org/compare-dl4j-torch7-pylearn


3. https://www.microway.com/hpc-tech-tips/deep-learning-framew]]></description><category></category></item><item><title><![CDATA[多语言神经机器翻译]]></title><link>https://blog.youkuaiyun.com/hellonlp/article/details/56495849</link><guid>https://blog.youkuaiyun.com/hellonlp/article/details/56495849</guid><author>hellonlp</author><pubDate>Wed, 22 Feb 2017 20:19:13 +0800</pubDate><description><![CDATA[1.  Multiple Language Neural Machine Translation
1.1 Multi-Task Learning for Multiple Language Translation (ACL 2015)
http://www.aclweb.org/anthology/P/P15/P15-1166.pdf



1.2 Multi-way, multili]]></description><category></category></item></channel></rss>