This project implement classic machine learning algorithms(ML). Motivations for this project includes:
- Helping machine learning freshman have a better and deeper understanding of the basic algorithms and models in this field.
- Providing the real-life and commercial executing methods in ML filed.
- Keeping my Mathematics Theory and Coding ability fresh due to such cases.
- Improve my ability of Java and Python.
Show how to use the package of fast_fm
to classify the training data directly.
@bolg:FM解析
We rewrite fm by ourselves and focus helping people get deeper insights about FM.So we upload it to the pypi named Fsfm
,you can downlode it if you're interested in it.
An interview problem in 'Nlp' solved by n-gram instead of Naive Bayes.
@bolg:SVD解析
@bolg:协同推荐解析
@bolg:评价文本判断用户流失倾向
@bolg:SMOTE解析
@bolg:风控方法解析
It means fast risk control with python.It's a lightweight tool that automatic recognize the outliers from a large data pool.
@bolg:Kaggle&TianChi分类问题相关算法快速实现
@bolg:Kaggle&TianChi分类问题相关纯算法理论剖析
T-distributed stochastic neighbor embedding(t-SNE) rewrite with Python by ourselves, it's a good dimensionality reduction method. Add many explanation among the code.
Some questions for the new hand to estimate their level of the ML、DL. What's more ,it also contains the key point which i think during my study with Andrew Ng's machine learning lessons(to be continued).
Also, I write some words to the new hand. Read it 写给想转行机器学习深度学习的同学 if you're interested in it .
Following the paper 'Deep Neural Networks for YouTube Recommendations' , finished with Python.
@bolg:利用DNN做推荐的实现过程中的总结
@bolg:关于'Deep Neural Networks for YouTube Recommendations'的一些思考和实现
See More From:
@bolg:基于Tensorflow实现FFM
More you may follow with interest :FM部分||deepFM部分
See More From:
@bolg:GolVe向量化做文本分类
More you may follow with interest :Youtube构造skn Vector||N-Grams
- Phone number analytical tools, design for get out the true phone number from digital mixed with dialect、chinese、special symbols
- Adjust that is any phone communication intention inside the conversation, base model coming from the result translated by IFLYTEK
pip install YMMNlpUtils==0.1.1
supported
download directly supported, here's the url: YMMNlpUtils 0.1.1
- 已知词扩展
- text_base = Neologism(st=text, prev_cut=True, macth_posseg=[["a"], ["n"]])
- text_base.filter(frequency=0.001, freedom=0.5)
- 新词发现
- direct_search = Neologism(st=text, prev_cut=False)
- direct_search.filter(frequency=0.0001, polymerization=15, freedom=0.5))
@bolg:Bayes_Optimizaion based on GP + UCB
- Java实现,Main为调用方式测试
- NLP文本重要性计算
- NLP文本去重
- SimiHash为Java实现
- Main为调用方式测试
- WordClassificationDeduplication为python版本
- Main为调用方式测试
- Java实现,Main为调用方式测试
- NLP文本重要性计算
- 关键词提取
- TextRankSummary用来做摘要提取,衍生自PageRank的迭代**
- TextRankKeyWord用来做关键词提取,衍生自PageRank的迭代**
- 关键词抽取,衍生自频率(freq)+由共现度得到的度(deg)的**,score = deg/freq,论文:Automatic Keyword Extraction from Individual Documents
- 优点:
- 快,算法简单而高效
- 能够提取一些较长的专业术语
- 缺点:
- 可以做召回,但是精确度欠佳
- 原论文基于英文,可切分词比较多,中文无法找到类似and ,or 这种切分词进行分段
- 实现:
- 平衡了高频词对全文的影响
- 采取了有效词长平衡,避免长文本造成的数据有偏现象
- 版本:
- 支持Java版本,依赖HanNlp分词器
- 支持Python版本,依赖jieba分词器
Python Environment. More details getting from single project requirement.
If you find some incorrect content, i'm so sorry about that. PLS contact me by the following way:
- WeChat:sharalion
- E-mail:stw386@sina.com
- Message Board in my bolg