Ten commandments of self-learning in machine learning

In the process of self-learning machine learning skills , We must be responsible for our own education and enlightenment . This article lists ten commandments that should not be taken lightly . warning taken from the overturned cart in front , Rear Division .

1. mathematics 、 Code and data are “ The trinity ” Of

Any proven machine learning assembly line , It's all math 、 The trinity of programming and data . Three aspects are equally important .

• If the data quality is not high , Then the mathematical formula is more beautiful , Code is efficient , It doesn't help .

• Even with high quality data , But if we don't know anything about Mathematics , Then it's going to be disappointing , It's even the result of a different direction .

• Even with world-class mathematical foundation and data , But in the face of inefficient programming code implementation , We still can't achieve large-scale benefits .

The data provide a mine rich in natural treasures , Mathematics is a mining tool , Programming provides a team of excavators .

mathematics 、 Programming and data “ The trinity ”, Make up the system input （ That is, the data on hand ） And the output （ The immediate result ） The bridge between .

Be careful ： Mathematics also includes the branches of statistics and probability theory , So you can put “ The trinity ” Think of it as a starfish .

2. Beyond the three , There are also specific needs to be considered

Besides thinking about how to strike a balance in the Trinity , It is also important to bear in mind that the ultimate target of the three services is customer demand . If you can't serve the customer , So even if there is code that performs the best 、 The most complete mathematical theory and the richest insight into data , It doesn't help .

Engineers are often obsessed with specific processes , And ignore the final output . Even if the starting point is better , It should not be forgotten that the actual results do not depend on the starting point .

Even if we don't offer good performance to our customers , It's better than solutions that don't give any performance at all .

for instance , If a state-of-the-art model needs to be run extra 47 Times of time , Finally, the accuracy is improved 1％ The effect of , Does this provide the best customer experience ？

3. Don't be “ The trinity ” Hoodwink

Even if we're very much in favor of the Trinity , But don't be blinded by your feelings .

Self taught machine learning Engineer , The biggest skeptic is himself .

They know that the data itself cannot confirm something , But it can be used to overturn something . for example , Only one outlier in a billion data points , It can prove that it is wrong to confirm an important idea before . Nature is not nonlinear , A flaw in Mathematics , There will be serious consequences . The performance of the code depends on its weakest point .

No matter how complete the Trinity is , You should not ignore your intuition . Once the result seems too perfect , Make you wonder if there might be something wrong with it , So, in fact, it is , Unless we're really lucky .

4. Maintain the relationship with potential customers

Let the machine do what it's good at , That is to achieve repetitive operations . People should do what they are good at , Including care 、 Empathy 、 question 、 listen 、 Leadership and teaching, etc .

Customers don't focus on math as much as we do 、 Programming and data , They care about themselves demand Is it satisfied .

When it comes to complex issues such as data ethics , Need from " Moral silver law "（Silver Rule） From the perspective of , namely “ What you don't want , Do not to others ”.

5. Respect the work of the former

When it comes to Computing Science 、 machine learning 、 Artificial intelligence 、 mathematics , We may immediately think of Ada Lovelace、Geoffrey Hinton、Yann LeCun、Yoshua Bengio、 Alan · Turing 、 Fei Fei Li 、Grace Hopper、 Wu enda 、 feng · Neumann 、Alan Kay、Stuart Russel、Peter Norvig Wait for the front runner .

Of course , Except for the familiar characters mentioned above , There are also thousands of people who have contributed that have not been remembered .

A budding rookie , You should be aware of the great contributions made by the pioneers . At the same time, we should realize that , For every new machine learner , Our predecessors all have the same expectations for us , It's that the future of this field depends on your current efforts .

6. The effect of overthrowing a new comeback should not be underestimated

Our goal is to provide reliable solutions in the first place . As your skills improve , We will Reexamine the original work , Break the old world , Rebuild from a new perspective .

For self-taught machine learning Engineers , It should be realized that, like nature , Software and machine learning programs are also ongoing , In constant development . Data is constantly changing 、 Code will also be executed on new hardware . also , A mathematical genius put forward what is called “Adam” The optimizer for , It is suitable for efficient computation of large data sets , It takes less memory resources .

We should be open to change , To welcome them . As soon as it changes , We should give the best judgment on whether it is suitable for implementation . Change is just something new , But that doesn't mean it's necessary .

7. Don't be a slave to tools

A common metaphor in the programming world is the problem of painting the bicycle shed . It refers to a programmer or team worrying about what color a bicycle shed should be painted , It's not about whether the shed can actually store bicycles .

Of course , Here bicycle shed corresponding to the specific purpose of the computer program .

In machine learning , We can hear a lot of endless arguments ,R still Python、TensorFlow still PyTorch、 Reading self-study or teaching in class 、 Math first or code first ,Spark still Hadoop、Amazon Web Services still Google Cloud Platform,VSCode still Jupyter、 NVIDIA or others, etc . in fact , There is no absolutely right choice .

All options apply , There's no need to argue between the top and the bottom .

The question we really need to answer is ： Which option is the fastest 、 The most reliable way to realize our idea ？

Once you think about it , We will find that all people are thinking about the same problem .

One of the biggest problems for Engineers , Is often from the tools to find problems , instead of Start by finding problems , And look for tools . In the face of problems , If you don't have the right tools , Then you need to build tools .

The same is true for educational resources . No matter how you study , mathematics , Code and data are immutable , The key is how to use them .

Keep in mind that , Many problems don't need the intervention of machine learning .

8. Ideas also have the value of communication

Don't confuse people who want to steal your ideas with people who want to follow your ideas . An idea may be much more valuable to others than to you .

As an engineer , Our role is not just to build our own ideas , And it's also about communicating with others , Explain how the idea can benefit others as well . If we don't have the ability to communicate , Then find partners who have or intend to promote their own communication skills .

In a society where authenticity is hard to tell , Sincerity will make us win . You should sincerely let others know what your work can offer , And what you don't know . It's an advantage to admit your own shortcomings , Not a disadvantage .

Good technology will remain invincible , That's not true of lies . Do more technology and less boast .

9. Compete with each other

See the progress of others , Are we jealous ？ Or see it as a potential work inspiration ？

How we feel about the success of others , It's also how we feel about our own success .

10. Don't be greedy

We try to be able to use mathematics 、 The ability of data and programming to provide solutions to target customers , But it should not be too demanding . Desire can lead to high expectations for the future , Instead of enjoying life now .

Keep your passion for learning , It is a good medicine to solve the problem of demanding skill improvement .

Learning engineers through self-taught machines , Can quickly master and control mathematics 、 Knowledge of programming and data , But don't be impatient . It needs to be clear , It takes time to learn to master any valuable technology . thus , The fun is in the process .

Once we start to teach ourselves , We need to fully undertake our own responsibility of enlightenment and education . In choosing a project, we should not hope to hit the big luck , We should choose carefully . Whether the project can meet their own quest ？ Will you challenge your existing skills ？ Whether you can follow the rules ？ If the answer is yes , That's enough .

And finally , The self-taught person should walk out a road suitable for himself , Standing on the tide forever ：

• Consolidate knowledge , Go for the certificate again .

• Do things in a down-to-earth manner , Don't overdo it .

• Teaching and learning .

• become an independent school .

• True knowledge comes from practice .

• Identify the purpose of each tool .

• Before delivery , Don't brag about .

• Have doubts , And put forward the hypothesis .

• one can't make bricks without straw .

• Based on the present , Planning for the long run .

All change is the same , The most important foundation of machine learning is mathematics 、 Programming and data .

Link to the original text ：

https://towardsdatascience.com/the-10-commandments-of-self-taught-machine-learning-engineers-9e810971ed34