Editor's note: This article is the fourth article in the "Planning for Recommended Course for Paradigm University," "How do ordinary programmers move closer to artificial intelligence? ", NetEase smart authorization starting.
I believe that friends who saw this article almost want to become machine learning scientists.
In fact, the vast majority of paid courses basically have completely free courses in another place. We just put this information together, telling you where to find them, and in what order to learn.
In this way, even if you are a college student who has not yet graduated, or an engineer who has just entered the workplace, you can master the basic skills of machine learning scientists through self-learning and apply it quickly in thesis, work, and even everyday life.
Here we recommend a user-friendly machine learning tutorial, you can become a machine learning scientist through months of learning, completely free.
A user-friendly machine learning tutorial
Are you overwhelmed by information overload when you learn machine learning courses?
Most of the learners have encountered this problem. This is not their fault because most machine learning courses pay too much attention to individual algorithms.燑/p>
Yes, although algorithms are important, they still spend too much time on algorithms.燑/p>
It's that... you can hardly walk through the process of machine learning in a short period of time and feel the excitement of solving specific data problems through it.
These machine learning courses focus on algorithms because it is easy to teach. In contrast, if the machine learning teacher is going to take you through the process of machine learning, then he needs to build the computing environment, complete the data collection, cleaning, splitting, feature processing, model adjustment and model prediction, and even he needs a Interactive interface for learners. There are so many tools for a teacher. It is better to learn machine learning algorithms instead of walking with the students.燑/p>
But the problem is that it is difficult for someone to stick to self-study and become an excellent machine learning scientist. Even if he is a doctor of mathematics, or a highly skilled programmer, it is easy to get caught in the details and it is difficult to have a sense of accomplishment with specific projects.燑/p>
This tutorial will bring completely different ideas. It is very suitable for self-study, even if there is no basis for programming, it can quickly implement machine learning models with appropriate tools to solve specific problems encountered in work and life.
It is worth noting that we have enjoyed the world's top machine learning resources without spending a penny.燑/p>
Way of self-learning 燑/inherit>
We recommend completing your study with Doing Shit (not a technical term).
Before you may have learned machine learning, but from the experience of my friends and I often get confused by various mysterious symbols, formulas, a large number of textbooks and papers, and then do not want to touch this annoying. Things.
Our approach will be more friendly, and its learning process will be like that of children. You will learn some basic knowledge (but not necessarily completely understand it) and then quickly implement it through easy-to-use tools. And when you are attracted by the results of modeling, then we talk about the mathematical logic and computational logic behind the algorithm.
So we will do a lot of machine learning projects in learning. The advantage of this is that when you face a job opportunity, you are an experienced machine learning scientist!
Of course, self-study is in need of self-discipline. This tutorial will always be with you. Here are 4 steps.
1. Preconditions (do not need to completely understand)
Statistics, programming, and math (you don't need programming) 燑/p>
2. Sponge mode 燑 / p>
Immerse yourself in various theories of machine learning.
3. Goal practice 燑/p>
Practice 9 interesting topics with machine learning packages.
4. Machine Learning Projects
Deep participation in projects and areas of interest
Step 1: Prerequisites 燑/b>
The reason why machine learning looks scary is because it is always accompanied by obscure terminology. In fact, even if you graduated from the Chinese Department, you can learn machine learning. However, we need you to have a basic understanding in some areas.
The good news is that once you meet the prerequisites, the rest will be very easy. In fact, almost all machine learning is applying the concepts of statistics and computer science to the data field.燑/p>
Task: Make sure you understand basic statistics, programming, and math.
Statistics: Understanding statistics, and Bayesian probability in particular, is critical to many machine learning algorithms.燑/p>
Free Guide:How to Learn Statistics for Data Science, The Self-Starter Way
Https://elitedatascience.com/learn-statistics-for-data-science
Programming: Understand that programming will be more flexible with applied machine learning.燑/p>
Free Guide:How to Learn Python for Data Science, The Self-Starter Way
Https://elitedatascience.com/learn-python-for-data-science
Mathematics: The study of original algorithms requires the foundation of linear algebra and multivariate calculations.燑/p>
Free Guide:How to Learn Math for Data Science, The Self-Starter Way
Https://elitedatascience.com/learn-math-for-data-science
You can take a look at these tutorials and lay the foundation for your machine learning path.燑/p>
Step 2: Sponge Mode 燑/b>
Sponge mode is to absorb as much knowledge of machine learning theory as possible.燑/p>
SpongeBob
Now some people may think, "If I don't plan to do original research, why do we still need to learn theory when we can use existing machine learning packages?"
This is a reasonable question!燑/p>
However, if you want to apply machine learning more flexibly to your day-to-day work, it's good to learn some basic theory, and you don't need to completely understand it. Below we will spoil the five reasons for learning machine learning theory.
(1) Planning and data collection
Data collection is an expensive and time consuming process! So what types of data do I need to collect? How much data do I need based on the model? Is this challenge feasible?燑/inherit>
(2) Data assumptions and preprocessing 燑/b>
Different algorithms have different assumptions about data input, so how do I preprocess my data? Should I regularize? If my model lacks some data, is it still stable? How to deal with outliers?燑/p>
(3) Interpretation of Model Results 燑/b>
Simply thinking that machine learning is a "black box" concept is wrong. Yes, not all results are directly explainable, but you need to diagnose your own models and improve them. How do I evaluate whether the model is overfit or underfit? How do I explain these results to business stakeholders? And how much room for improvement does the model have?燑/p>
(4) Improve and adjust model 燑/b>
Your first training rarely reaches the best mode. You need to understand the nuances of different parameters and regularization methods. If my model is overfitting, how can I remedy it? Should I spend more time on feature engineering or data collection? Can I combine my models?燑/p>
(5) Driving business value 燑/b>
Machine learning is never done in a vacuum. If you do not understand the tools in the arsenal, you cannot maximize their effectiveness. Among the many outcome indicators, which are the optimized reference indicators? Which is more important? Or are there other algorithms that will perform better?燑/p>
The good news is that you don't need to know the answers to all questions from the beginning. So we recommend that you start with learning enough theory and then quickly move into practice. In this case, you are more able to stick with it, and after a certain period of time you are really good at machine learning.
Here are some free machine learning materials.
2.1 Machine Learning Video Courses 燑/b>
This is a world class course from Harvard University and Yale University.
Task: Complete at least one course
Harvard University Data Science Courses
End-to-end data science courses. Compared to Wu Enda's curriculum, it places less emphasis on machine learning, but from data collection to analysis, you can learn the entire data science workflow here.燑/inherit>
Course homepage: http://cs109.github.io/2015/
Stanford University Machine Learning Courses
This is Wu Nida's famous course. These videos clearly explain the core concepts behind machine learning. If you can only take one class in your time, we recommend this one.燑/p>
Course homepage: https://?v=qeHZOdmJvFU&list=PLZ9qNFMHZ-A4rycgrgOYma6zxF4BZGGPW&index=1
2.2 Machine Learning References 燑/b>
Next we recommend two classic textbooks for the industry.燑/p>
Task: Look at these PDFs as textbooks.
An Introduction to Statistical Learning
Gentler introduced the basic elements of statistical learning in the book and is suitable for all machine learning learners.燑/p>
PDF Address: http://~gareth/ISL/ISLR%20Sixth%20Printing.pdf
Elements of Statistical Learning燑/p>
Strictly introduced machine learning theory and mathematics, recommended to machine learning researchers.燑/p>
PDF Address: http://statweb.stanford.edu/~tibs/ElemStatLearn/
2.3 The key to success
The following is the key to the success of each step.燑/p>
A: Focus on the big picture, always ask why 燑/p>
Whenever you are introduced to a new concept, ask "why." Why use decision trees instead of regressions in certain situations? Why regulate parameters? Why split the data set? When you understand why you use each tool, you will become a true machine learning practitioner.燑/p>
B: Accepting you won't remember everything you learned.
Don't take crazy notes, and don't review every lesson 3 times. In your actual work, you will often need to look back.
C: Keep going, don't be discouraged.
Try to avoid long time on one topic. Even for machine learning professors, some concepts are very difficult to explain. But when you begin to use it in practice, you will soon understand the true meaning of the concept.燑/p>
D: Video is more effective than textbooks.
From our experience, textbooks are a good reference tool, but it is difficult to adhere to. We strongly recommend video lectures.燑/p>
Step 3: Purposeful practice 燑/inherit>
After the sponge mode, we will hone our skills through deliberate exercises to raise the level of machine learning to a new level. The goal includes three aspects:
1. Practice a complete machine learning process: including data collection, cleaning, preprocessing, model building, tuning parameters and model evaluation.燑/p>
2. Practice in real data sets and gradually establish which model fits the instinct of what kind of challenge.
3. Go deep into a specific topic, such as applying different types of clustering algorithms in the data set to see which works best.燑/p>
After completing these steps, you will not be at a loss when you begin to solve large projects.
3.1 Machine Learning Tools in/inherit>
In order to quickly implement a machine learning model, we recommend using an off-the-shelf modeling tool. In this way, you will practice the entire machine learning workflow in a short period of time without spending too much time on any one step. This will give you very valuable "Big Picture Intuition".
Python: Scikit-Learn燑/p>
Scikit-learn and Sklearn are the gold standard libraries for Python in general machine learning, and have implementations of conventional algorithms.燑/p>
R:Caret燑/p>
Caret provides a uniform interface for model packages in the R language. It also includes preprocessing, data splitting, and model evaluation capabilities, making it a complete end-to-end solution.燑/p>
3.2 practice data set 燑/inherit>
After learning the tools, you also need some data sets. The art of data science and machine learning, many of which are in the dozens of micro-level decisions when solving problems. We will see the results of the modeling in different data sets.
Task: Select 5 to 10 data sets from the following options. We recommend starting with UCI's machine learning library. For example, you can select 3 data sets for regression, classification, and clustering.燑/p>
When doing machine learning projects, think about the following questions:
What kind of preprocessing do you need to perform for each dataset? Do you need to reduce dimension? What method can you use?燑/li> How can you split the data set?燑/li> How do you know if the model has "overfitting"?哪些/li> What types of performance indicators should you use?燑/li> How do different parameter adjustments affect the results of the model?燑/li> Can you combine models to get better results?燑/li> Does your clustering result match intuitively?燑/li>UCI Machine Learning Reports
The UCI Machine Learning Report collects over 350 different data sets and provides training data specifically for machine learning. You can search by task (regression, classification, or clustering) or search by industry, dataset size.燑/p>
Address: http://archive.ics.uci.edu/ml/
Kaggle燑/p>
Kaggle.com is an enthusiastic apologetic. There are 180 community datasets that contain interesting topics. There is everything from user Pokemon to European football matches. .燑/p>
Https://
Data.gov燑/p>
If you are looking for social science or government-related datasets, check out Data.gov. This is a collection of open data from the U.S. government. You can search more than 190,000 data sets.燑/p>
Https://
Step 4: Machine Learning Projects 燑/b>
OK, now it's really interesting. So far, we have covered preconditions, basic theories, and purposeful practices. Now we are ready to enter a bigger project.燑/p>
The goal of this step is to integrate machine learning techniques into a complete, end-to-end analysis.燑/p>
4.1 Complete a Machine Learning Project
Mission: Complete the Titanic Survivor Challenge.
The Titanic Survivor Prediction Challenge is a very popular machine learning practice project. In fact, this is a ggleaggle.com contributing team.
We like to use this project as a starting point because it has many great tutorials. You can learn how these experienced data scientists deal with data exploration, feature engineering, and model tuning.燑/p>
Python tutorials
We really like this tutorial because it teaches you how to preprocess the data and correct the data. Tutorial provided by Pycon UK.燑/p>
Tutorial address: https://github.com/savarin/pyconuk-introtutorial
R tutorial
Use the Caret package in R to handle several different models. This tutorial is a good summary of the end-to-end predictive modeling process.燑/p>
Tutorial address: http://amunategui.github.io/binary-outcome-modeling/
This is an "irresponsible" quick tutorial: just a tutorial, skipping the theoretical explanation. However, this is also useful, and it shows how to conduct random forest operations.燑/p>
Tutorial address: http://will-stanton.com/machine-learning-with-r-an-irresponsibly-fast-tutorial/
4.2 to write an algorithm from scratch 燑/inherit>
In order to have a deeper understanding of machine learning, nothing is more helpful than writing an algorithm from scratch because the devil is always in the details.
We recommend starting with some simple things, such as logistic regression, decision trees, or KNN algorithms.燑/p>
This project also provides you with a practice of translating data languages ​​into programming languages. This skill will be very convenient when you want to apply the latest academic research to your work.燑/p>
And if you get stuck, here are some tips:
Wikipedia has a lot of good resources. It has many pseudocodes for common algorithms. To nurture your inspiration, try to view the source code of an existing machine learning package. Decompose your algorithm and write separate functions for sampling, gradient descent, etc. Start simple and execute a decision tree before attempting to write a random forest.4.3 Choose an interesting project or area 燑/inherit>
If you are not curious, you are hard to learn. But so far, maybe you have found a field that you want to stick to, so start modeling!
To be honest, this is the best part of machine learning. This is a powerful tool, and as soon as you begin to understand, many ideas come to you.燑/p>
The good news is that if you've been tracking and you're ready for the job, your gains will be far beyond your imagination!燑/p>
We also recommended 6 interesting machine learning projects.燑/p>
Address: https://elitedatascience.com/machine-learning-projects-for-beginners
Congratulations on your arrival at the end of the self-study guide
The good news here is that if you have followed and completed all tasks, then you will be better at applying machine learning than 90% of people who claim to be data scientists.燑/p>
And the better news is that you still have a lot to learn. For example, deep learning, reinforcement learning, migration learning, confrontation generation models, and so on.燑/p>
The key to becoming the best machine learning scientist is to never stop learning. Start your journey in this vibrant and exciting area!
Note: This tutorial was provided by EliteDataScience. We translated this tutorial with a slight change. Original link: https://elitedatascience.com/learn-machine-learning
About Paradigm University:
"Pattern University" was initiated by the fourth paradigm and is dedicated to becoming a "Data Scientist" Whampoa Military Academy. The "Pattern University Series" will share with the top practitioners such as Dai Wenyuan, Yang Qiang, and Chen Yuqiang in the field of machine learning, as well as machine learning materials recommended and organized by the fourth paradigm product team.
燑br>
Organize your cabling
- Organizes all types of cables
- Ideal for network Patch Panel cabling
- Finger duct design allows easy cable organization
- Cable pass-through holes in the back of the duct
- Only set up 1U rack
- Work with threaded, round and square hole racks
- EIA 19" RACK standard compliable
- Including screws and mounting hardware
Metal Cable Management,1U Metal cable management,stainless steel cable management,metal cable tie mount
NINGBO UONICORE ELECTRONICS CO., LTD , https://www.uonicore.com