Wu Enda's machine learning mind mapping insists on clocking in for 23 days - building a knowledge context, reviewing, summarizing and replying

Author：AXYZdong Automation Engineering Male
A little bit of thinking , Have a little idea , A little bit rational ！
Set a small goal , Try to be a habit ！ Meet a better self in the most beautiful years ！
[email protected],CSDN First episode ,AXYZdong original
The only blog update address is ： AXYZdong The blog of
B The homepage of the website is ：AXYZdong My personal homepage

0. Preface

Machine learning is one of the most exciting directions in information technology . This paper takes teacher Wu Enda's machine learning course as the main line , Use Process On Online drawing constructs the mind map of machine learning .

1. Instructions for using mind map

Cooperate with teacher Wu Enda's machine learning video , Build knowledge context , Review and summarize the reply .

Browse all mind maps online ： Wu Enda machine learning - Mind Mapping ProcessOn

Students who need to browse the full picture pay attention to AXYZdong official account , reply machine learning Get the password ！

---

2. The main content of mind mapping

introduction （Introduction）

Supervised learning part ：

1. Univariate linear regression （Linear Regression with One Variable）
2. Multivariate linear regression （Linear Regression with Multiple Variables）
3. Logical regression （Logistic Regression）
4. Regularization （Regularization）
5. neural network ： describe （Neural Networks:Representation）
6. neural network ： Study （Neural Networks:Learning）
7. Support vector machine （Support Vector Machines）

Unsupervised learning part ：

1. clustering （Clustering）
2. Dimension reduction （Dimensionality）
3. Anomaly detection （Anomaly Detection）

Special applications ：

1. Recommendation system （Recommender Systems）
2. Large scale machine learning （Large Scale Machine Learning）

Suggestions on establishing machine learning system ：

1. Suggestions for applied machine learning （Advice for Applying Machine Learning）
2. The design of machine learning system （Machine Learning System Design）
3. Application example ： Picture text recognition （Application Example: Photo OCR）

---

3. Mind map text

0. introduction （Introduction）

The introduction mainly introduces the definition of machine learning 、 Related algorithms of machine learning 、 Supervised learning and unsupervised learning .

There is no uniform definition of machine learning , The following two are the two scholars' understanding of machine learning mentioned in the video .

• Arthur Samuel (1959). Machine Learning: Field of study that gives computers the ability to learn without being explicitly programmed.
• Tom Mitchell (1998). Well-posed Learning Problem: A computer program is said to learn from experience E with respect to some task Tand some performance measure P, if its performance on T, as measured by P, improveswith experience E.

^{▲ Part0 Introduction}

1. Univariate linear regression （Linear Regression with One Variable）

The main content of this part includes the model representation of univariate linear regression 、 Cost function 、 Gradient descent method and using gradient descent method to solve the minimum value of cost function .
​

^{▲ Part1 Linear Regression with One Variable}

2. Multivariate linear regression （Linear Regression with Multiple Variables）

Multivariable linear regression is equivalent to the expansion of univariate , Mainly according to the model assumptions 、 The idea of constructing cost function and studying the minimum value of cost function .

Unlike univariate linear regression , Multivariate linear regression may also involve the problem of feature scaling , The main reason is that there are characteristic variables with different scales , In order to make the gradient descent converge quickly , These characteristic variables need to be unified （ Similar to the idea of normalization ）

Compared with univariate linear regression , Multivariable linear regression is used to solve the characteristic equation of cost function , In addition to the gradient descent method , You can also use regular equations . According to the number of characteristic variables , Choose these two methods flexibly .
​

^{▲ Part2 Linear Regression with Multiple Variables}

3. Logical regression （Logistic Regression）

there “ Return to ” Different from linear regression , It's a customary name . Its essence is classification , The variables to be predicted are discrete values .

^{▲ Part3 Logistic Regression}

4. Regularization （Regularization）

Regularization （Regularization） The proposed , It mainly solves the problem of fitting （over-fitting） The problem of . Including the regularization of linear regression and the regularization of logical regression , Its essence is to preserve all features by adding regularization terms , At the same time, reduce the parameters （ Coefficient before characteristic variable ） Size .

One hypothesis can be better fitted to the training data than others , But it can't fit the data well on the data set other than the training data , At this time, we think that there is a phenomenon of over fitting in this hypothesis . The main reasons for this are noise or too little training data .
​

^{▲ Part4 Regularization}

5. neural network ： describe （Neural Networks:Representation）

neural network （Neural Networks） A brief statement of , Involving nonlinear assumptions 、 Model representation of neural networks 、 Intuitive understanding of neural networks and multiple classification .

When there are too many features , Ordinary logistic regression model , Can't handle so many features effectively , Now we need neural networks .
​

^{▲ Part5 Neural Networks:Representation}

6. neural network ： Study （Neural Networks:Learning）

neural network （Neural Networks） The cost function of , Gradient descent seeks the minimum value of the cost function , Using back propagation algorithm （Backpropagation Algorithm） Calculate the direction of gradient descent .

Numerical tests using gradients （Numerical Gradient Checking） Method , The cost of prevention seems to be decreasing , But the final result may not be the problem of optimal solution .

If you let the initial parameters be 0, Then the activation unit of the second layer will have the same value . Therefore, you need to initialize the parameters , The method of random initialization is adopted ,Python The code is as follows ：

Theta1 = rand(10,11) * (2*eps) - eps

^{▲ Part6 Neural Networks:Learning Learning}

7. Suggestions for applied machine learning （Advice for Applying Machine Learning）

• When using the trained model to predict unknown data, it is found that there is a large error , What to do next ？ Use the diagnostic method to judge which methods are effective for our algorithm .

• The training set and test set are used to evaluate whether the hypothesis function is over fitted , The parameters obtained by minimizing the cost function of the training set are substituted into the cost function of the test set .

• Cross validation sets to help select models . Diagnostic bias and variance , The performance of the algorithm is not ideal , Or the deviation is relatively large , Or the variance is bigger . let me put it another way , What happens is either an under fit , It's either an over fitting problem .

• The learning curve takes the training set error and cross validation set error as the number of training set instances （m） The graph drawn by the function of .
  ​
  

  ^{▲ Part7 Advice for Applying Machine Learning}

8. The design of machine learning system （Machine Learning System Design）

The main content of this part is error analysis 、 The error measure of class skew 、 The trade-off between precision and recall and machine learning data .
​

^{▲ Part8 Machine Learning System Design}

9. Support vector machine （Support Vector Machines）

Support vector machine (Support Vector Machines) In essence, it is to optimize the objective function in logistic regression , Will contain log Item usage cost Function instead of .

Support vector machine uses a maximum spacing to separate samples , Robust , It is sometimes called a large spacing classifier .

Kernel function (Kernel) Introduce support vector machine SVM in , Instead of the corresponding high-dimensional vector inner product .
​

^{▲ Part9 Support Vector Machines}

10. clustering （Clustering）

clustering （Clustering） One kind of unsupervised learning .

Key algorithms ：K- Mean algorithm .K-Means Is the most popular clustering algorithm , The algorithm accepts an unmarked data set , Then clustering the data into different groups .

^{▲ Part10 Clustering}

11. Dimension reduction （Dimensionality）

Dimension reduction （Dimensionality） It is mainly used for data compression and data visualization , It is also a kind of unsupervised learning .

Important algorithm ： Principal component analysis PAC（Principal Component Analysis） Algorithm .

^{▲ Part11 Dimensionality}

12. Anomaly detection （Anomaly Detection）

This part mainly includes Gaussian distribution （Gaussian Distribution）, Gaussian algorithm is used for anomaly detection , Feature transformation transforms the original data into Gaussian distribution .

Important algorithm ： gaussian （Gaussian ） Algorithm .

^{▲ Part12 Anomaly Detection}

13. Recommendation system （Recommender Systems）

This part includes ： Content based recommendation system 、 Collaborative filtering （Collaborative Filtering）、 Vectorization ： Low rank matrix decomposition 、 Implementation details ： Mean normalization .

Important algorithm ： Collaborative filtering （Collaborative Filtering） Algorithm .

^{▲ Part13 Recommender Systems}

14. Large scale machine learning （Large Scale Machine Learning）

primary coverage ： Random gradient descent method （Stochastic Gradient Descent）、 Small batch gradient descent （Mini-Batch Gradient Descent） 、 Convergence of stochastic gradient descent algorithm 、 Online learning （Online Learning） and Mapping simplification and data parallelism （Map Reduce and Data Parallelism）.

^{▲ Part14 Large Scale Machine Learning}

15. Application example ： Picture text recognition （Application Example: Photo OCR）

Focus on Steps of image and character recognition and The sliding window （Sliding Windows） Use .

^{▲ Part15 Application Example: Photo OCR}

16. summary （Conclusion）

^{▲ Part16 Conclusion}

4. About in the title “ Keep punching 23 God ”

Blink Keep punching 23 God

^{▲ Blink Clock in 23 God}

5. reference

[1]：[ Chinese and English subtitles ] Wu Enda machine learning series
[2]：fengdu78, Coursera-ML-AndrewNg-Notes, (2018), GitHub repository, https://github.com/fengdu78/Coursera-ML-AndrewNg-Notes

   This sharing is here

---

If my blog helps you 、 If you like my blog content , please “ give the thumbs-up ” “ Collection ” “ Focus on ” One button, three links ！

For more, please go to AXYZdong The blog of

---

If any of the above is inaccurate , Welcome below Leave a message . Or you have a better idea , Welcome to exchange and study together ~~~