This is one of the biggest mistakes beginners make.
1. Many beginners think that if you directly use an advanced API like TensorFlow, it is called a library adjuster, which seems to be low-level.
2. It seems that if you don't realize the research of algorithm theory and mathematical calculation from scratch, you are learning fake knowledge.
But this is the reason for the mistake, people are too susceptible to the influence of the recruitment market, don't you know that the recruitment market wants to spend the money of a delivery man to recruit an academician of the Chinese Academy of Sciences, if you don't want money, it will be more perfect.
So in retrospect, you don't have to implement a machine learning algorithm from scratch.
First of all, there are countless algorithms in various fields, and even if you want to eat the transparent white principle for an algorithm in a very subdivided field, it may take a few years. Second, this learning style is still the concept of the bottom-up approach traditionally used to teach machine learning.
Learn math. Learning Theory.
Implement the algorithm from scratch.
I sublimated it by this step.
Apply machine learning.
Algorithm implementation traps that beginners tend to fall into. Designing world-class machine learning algorithm implementations is very difficult. So you should use an off-the-shelf implementation.
More complex nonlinear methods require more data than linear methods. Algorithms need to process all this data quickly. This may require reinterpreting the linear algebra behind the method in a way that best suits the specific matrix operations in the underlying library. It may require specialized caching knowledge to get the most out of your hardware.
These are not things that can be solved once the "hello world" implementation works.
1. Don't hide in the mountains and practice hard, apply what you have learned as soon as possible, explore deeper knowledge in learning and project practice, or improve your mathematical skills.
2. Learn to identify what kind of knowledge needs to be investigated, what kind of knowledge is not needed, time is limited, and life is short.
3. Pay attention to coding, and coding ability will greatly affect your learning efficiency.
4. Don't reinvent the wheel, look more, and ask yourself if it is necessary to write from scratchCan you find it by reusing existing open-source algorithms in libraries or tools?
I'm a tech creator