I help beginners get started with machine learning. But time and time again the same mistakes are seen in mentality and action.
In this post, you'll find what I think are the 5 most common ways beginners make mistakes when starting machine learning.
I firmly believe that anyone can get started, use applied machine learning and do a great job. Hopefully, you'll recognize yourself in one or more of the traps below and take some corrective action to get back on track.
The traditional machine learning teaching method is bottom-up.
Study hard with a math background.
Work hard to learn the theory of machine learning.
Work to implement the algorithm from scratch.
Confused stage.
Finally start using machine learning (your goal!) ）。
This approach is slow and difficult (designed for academics who want to extend the latest technology, not for practitioners who want results)
pitfall
If you want to or say the following, you know you're in this trap:
I need to complete this linear algebra course first.
I need to go back and study for my PhD. primary.
I had to read this textbook first.
suggestion
How does studying 4 years of math or esoteric algorithmic theory get you to your desired goal? You're more likely to stop and fail.
The workaround is to flip the model.
If the valuable contribution of machine learning to the market is a set of accurate predictions, learn how to model problems and make accurate predictions. Start here.
And then really good at using it.
Read, steal, exploit theory if needed, but only for your goals. Only if it makes you better at delivering value.
Machine learning is a very large area of research.
It is the automation of the computer learning process and has a deep overlap with artificial intelligence.
From esoteric learning theories to robotics. The field is enormous.
The field is too big for you to afford all.
If you have the following thoughts, you know you have succumbed to this trap:
I need to know about every new technology mentioned on the new website.
I need to learn computer vision, natural language processing, speech, etc. first.
I need to know everything.
Choose a small corner and focus on it. Then narrow the scope again.
The most valuable area of machine learning is predictive modeling. Create a model from your data to make predictions.
Next, focus on one of the most relevant or interesting predictive models that are most relevant to you.
Then stick with it.
Maybe you're choosing through technology, like deep learning. Or you can choose by question type, such as a referral system.
Maybe you're not sure, so pick one. Become excellent or at least proficient.
Then, later, go back to another area.
Machine learning is actually about algorithms.
There are a lot of algorithms. Each algorithm is a complex system, and it has its own small area of study. It has its own ecosystem.
People lost in algorithms are called scholars, and this is not the goal of the masses.
If you find yourself saying:
Before I can use it, I need to know why it works.
I need to dive into hyperparameters first.
I need to explain the causes and effects when adjusting.
The algorithm is not the result. They are the means to an end.
In fact, machine learning algorithms are a commodity.
Swap them out. Try a number of methods for your problem. Make some adjustments to them, but move on.
You can learn more about the algorithm for better results, but know when to stop.
Procedures for using the system. Design tuned experiments and automate execution and analysis.
Machine learning is about good use of algorithms, but applying machine learning is more than just fiddling with algorithms.
Focus on the goal of delivering results from each project, i.e. a set of predictions or models that can achieve them.
You can learn a lot from implementing an algorithm from scratch.
Sometimes you even need to implement a technology because there is no suitable or usable implementation.
However, more often than not, you don't have to and shouldn't.
You fall into this trap if:
You're writing code to load the csv file (what the hell!?). ）
You are writing code for standard algorithms such as linear regression.
You are writing code for cross-validation or hyperparameter tuning.
Stop it.
Use a common library used by tens of thousands or hundreds of thousands of other developers that handles all edge situations and is known to be correct. Use a highly optimized library to compress each last cycle and each last byte of memory from the hardware. Use a graphical user interface for your own projects and avoid code altogether. Implementing it every time you want to use it is a very slow way to get started with machine learning.
If you're implementing for learning, be honest with yourself and separate it from learning how to deliver value through applied machine learning.
There are a lot of great machine learning tools.
In fact, great tools, data availability, and fast hardware are the reasons we're seeing a renaissance in machine learning.
However, you may fall into the trap of jumping to every new tool you stumble upon.
If you find yourself, you fall into this trap:
Use every new tool you've heard about.
Find yourself learning a new tool or language every week or month.
Learn a library and leave it behind to go to a new library.
Learn and use new tools.
But be strategic.
Integrate new tools into your systematic processes for solving machine learning problems.
If you choose one of the big major platforms and stick with it, you'll be more productive in solving problems, at least until you're proficient or proficient in it.
There are other tools, and if this is your field, there are many more professional tools.
Following in the footsteps of others is the difference between amateurs and professionals.
#Machine Learning##Data Analytics##Artificial Intelligence##工作方法 #