首页
学习
活动
专区
工具
TVP
发布
精选内容/技术社群/优惠产品,尽在小程序
立即前往

Summary of Chap2,End-to-End Machine Learning Project

This article is part of a series of summaries on the book Hands-On Machine Learning with Scikit-Learn and TensorFlow.

Personally, I split the understanding of machine learning into 3 levels.

Just for understanding - Just the concept and idea

Techician - Use of library and tools to build machine learning project

Scientist/Engineer - Understand the maths and statistics behind the models.

This article summarizes the machine learning process in plain language as much as posible. It covers the general concepts and gives you rough ideas of how does machine learning work, leaving the deep stuffs untouched. So this summary is great for the first level, also serves a good start of the next two levels.

For details and source code, please refer to the reference. Based on individuals background and time available, there are various versions to you can choose to read,1-minute-summary,10-minutes-summary.

1-minute-summary

If you want to conclude machine learning project within 5 steps, here is how I do it.

Discover and visualize the data to gain insights.

Prepare the data for Machine Learning algorithms.

Select a model and train it.

Fine-tune your model.

As you can probably tell, training a model is not the main part of machine learning. In fact, even though building machine learning models is the core, it is only a smart part of the whole machine learning project. I had this mis-understanding before I read this book. That's also the most important lesson I learn from Chapter 2.

10-minutes summary

First step, understand your data

It is important to understand your data. And the most intuitive way is to visualize it.

For individual features

Use histogram to plot distribution of individual features. The distribution can be useful. For example, you can obtain the support(range) of the feature. Sometimes, it is important to stratify the data and make sure each class has approximately same amount of training data.

Histogram

Feature correlations

A quick and efficient way to understand how "related" a feature with the goal is to draw the coorelations.

Correlation

Geographical

If the data has geographical or geometric nature, plotting against the nature shape of the data will be very helpful too.

Geographical

Secondly, prepare the data

Just like a chef needs to prepare the food before cooking, a data analyst needs to prepare the data before modeling. Common approaches include:

(1) fill missing values in raw data

(2) derive additional features

(3) categorize data

(4) standardization and normalization

(4) others

Next, modeling

Try many models

Most of the machine learning problems can be treated as either classification problems or regression problems.

There are many models for each category. For example, for classification problem, there areSVM,Decision Tree,Random Forest,Logistic Regression,Native Bayes, etc. Based on the nature of the problem and the data, some models fit better, some poor. So try out multiple models to your problem and select the few that fit better.

Prediction result can be improved by "ensemble method", which will be discussed in the future chapters.

Model Evaluation

The accurary of a model can be evaluated using a technique called cross validation.

Intuitively, a model of high accuracy seems to be better. That's a good guess, but not a complete answer. Based on the nature of the problems, especially the cost of wrong predictions, accuracy is not the only benchmark.

Last but not least, Fine-tune the model, and Launch

There are probably a few hyper-parameters in your models. Certain methods like randomized search can be used to find the optimal parameters. Also, ensemble methods may give you better overall results.

It is fun to try and build models. And it is more meaningful to finally put it in the real environment. However, you may need to regularly re-train your model.

Reference

I plan to read carefully through the bookHands-On Machine Learning with Scikit-Learn and TensorFlowfrom O'Reilly and follows the examples in it. AndI will share my summaries on the chapters together with some of my own understanding. The source codes are from either the book or this GitHub repository https://github.com/ageron/handson-ml.

This book is strongly recommended for machine learning beginners. It introduces the machine learning with two of the most popular machine learning tools,TensorFlowandScikit-Learn. What's more, it does not require deep maths and programming knowledge as pre-requisites. All you need to know is the very basic of Python.

P.S.1

寻一公众号合伙人,这样既可提供更多的素材,又可以相互校稿,翻译,相互学习进步,更能给读者提高更好的服务。如果有公众号维护经验是最好不过。

P.S.2

由于小编是根据着本书的英文版做的总结,对于一些术语的中文名不是很确定,所以不敢提供中文版总结误人子弟。如果某位领域的先行者可以热心帮忙提供中文术语的校正,请联系小刘,感激不尽。

  • 发表于:
  • 原文链接http://kuaibao.qq.com/s/20180320G1U4UX00?refer=cp_1026
  • 腾讯「腾讯云开发者社区」是腾讯内容开放平台帐号(企鹅号)传播渠道之一,根据《腾讯内容开放平台服务协议》转载发布内容。
  • 如有侵权,请联系 cloudcommunity@tencent.com 删除。

扫码

添加站长 进交流群

领取专属 10元无门槛券

私享最新 技术干货

扫码加入开发者社群
领券