Shulou(Shulou.com)06/01 Report--

This article introduces the relevant knowledge of "what is machine learning". In the operation of actual cases, many people will encounter such a dilemma. Then let the editor lead you to learn how to deal with these situations. I hope you can read it carefully and be able to achieve something!

What is machine learning? ML

The concept of machine learning holds that you don't need to write any special program code for the unsolved problem, and the generic algorithm (generic algorithms) can get interesting answers for you on the data set. For generic algorithms, instead of coding, the data is entered, and it builds its own logic on top of the data.

For example, there is a class of algorithms called classification algorithms, which can divide data into different groups. A classification algorithm used to identify handwritten numbers can be used to divide e-mail into spam and ordinary mail without modifying a single line of code. The algorithm has not changed, but the input training data has changed, so it has different classification logic.

Machine learning algorithm is a black box and can be reused to solve many different classification problems.

"Machine learning" is an inclusive term that covers a large number of similar generic algorithms.

Two kinds of machine learning algorithms

You can think of machine learning algorithms as two categories: supervised learning (Supervised Learning) and unsupervised learning (Unsupervised Learning). The difference between the two is simple, but very important.

Supervised learning

Suppose you are a real estate agent and the business is getting bigger and bigger, so you hire a group of interns to help you. But here's the problem-you can take a look at the house and know how much it's worth, and the interns are inexperienced and don't know how to value it.

To help your interns (perhaps to liberate yourself for a vacation), you decide to write a little software that assesses the value of houses in your area based on factors such as the size of the house, the location, and the transaction price of similar houses.

You wrote down every housing transaction in the city over the past three months, and each one you recorded a long list of details-the number of bedrooms, the size of the house, the location, and so on. But most importantly, you wrote down the final transaction price:

This is our "training data".

This is called supervised learning. You already know the price of each house, in other words, you know the answer to the question, and you can reverse the logic of solving the problem.

In order to write software, you will include training data for each property into your machine learning algorithm. The algorithm tries to figure out what operation should be used to get the price number.

It's like an arithmetic exercise, where all the symbols in the formula are erased:

Even if you don't want to predict unknown data (such as prices), you can use machine learning to do something interesting.

It's kind of like someone giving you a piece of paper with a lot of numbers listed on it and saying to you, "I don't know what these numbers mean. Maybe you can find patterns or sort them out, or something-good luck!"

What should you do with the data? First of all, you can use an algorithm to automatically divide different market segments from the data. You may find that buyers near the university prefer small houses with many bedrooms, while suburban buyers prefer large three-bedroom apartments. This information can directly help your marketing.

You can also do a cool thing and automatically find out the outliers of house prices, that is, values that are very different from other data. These outstanding properties may be high-rise buildings, and you can concentrate your best salesmen in these areas because their commissions are higher.

In the rest of this article, we mainly discuss supervised learning, but this is not because unsupervised learning is not useful or boring. In fact, unsupervised learning is becoming more and more important as the algorithm improves so that there is no need to associate data with the correct answer.

There are many other kinds of machine learning algorithms. But it was a good understanding when I was a beginner.

That's cool, but can evaluating house prices really be seen as "learning"?

As a human being, your brain can cope with most situations and can learn how to deal with them without any specific instructions. If you have been a real estate agent for a long time, you will have an instinctive "feeling" about the right pricing of the property, the best way to market it and which customers will be interested. The goal of strong artificial intelligence (Strong AI) research is to be able to replicate this ability with computers.

But current machine learning algorithms are not that good-they can only focus on very specific and limited problems. Perhaps in this case, the more appropriate definition of "learning" is to "find an equation to solve a particular problem based on a small amount of sample data."

Unfortunately, the name "machines come up with an equation to solve a particular problem based on a small amount of sample data" sucks. So finally we replaced it with "machine learning".

Of course, if you read this article 50 years later, we have already come up with strong artificial intelligence algorithms, and this article looks like an antique. Human of the future, you'd better not read it and ask your robot servant to make you a sandwich.

Let's write the code!

How do you plan to write the procedure for evaluating house prices in the previous example? Think about it before you look down.

If you know nothing about machine learning, chances are you will try to write some basic rules to assess house prices, as follows:

Python

Def estimate_house_sales_price (num_of_bedrooms, sqft, neighborhood): price = 0 # In my area The average house costs $200 per sqft price_per_sqft = 200 if neighborhood = = "hipsterton": # but some areas cost a bit more price_per_sqft = 400 elif neighborhood = = "skid row": # and some areas cost less price_per_sqft = 100 # start with a base price estimate based on how big the place is price = price_per_sqft * sqft # now adjust our estimate based on the number of bedrooms if num_of_bedrooms = = 0: # Studio apartments are cheap price = price   -  20000 else: # places with more bedrooms are usually # more valuable price = price + (num_of_bedrooms * 1000) return price

If you fool around like this for a few hours, you may achieve some results, but your program will never be perfect and will be difficult to maintain when prices change.

Wouldn't it be better if the computer could find a way to achieve the above functions? As long as the house price figures returned are correct, who cares what the function does?

Def estimate_house_sales_price (num_of_bedrooms, sqft, neighborhood): price = return price

One way to think about this is to think of house prices as a bowl of delicious soup, which consists of the number of bedrooms, size and location. If you can figure out how much impact each ingredient has on the final price, you may be able to get the specific proportion of the ingredients that mix together to form the final price.

This simplifies your original program (all crazy if else statements) to something like this:

Def estimate_house_sales_price (num_of_bedrooms, sqft, neighborhood): price = 0 # a little pinch of this price + = num_of_bedrooms * .841231951398213 # and a big pinch of that price + = sqft * 1231.1231231 # maybe a handful of this price + = neighborhood * 2.3242341421 # and finally, just a little extra salt for good measure price + = 201.23432095 return price

Pay attention to the magic numbers in bold-.841231951398213, 1231.1231231231, 2.3242341421, and 201.23432095 They are called weights. If we can find the perfect weight for each house, our function can predict all house prices!

A stupid way to find the best weight is as follows:

Step 1:

First, set each weight to 1.0:

Def estimate_house_sales_price (num_of_bedrooms, sqft, neighborhood): price = 0 # a little pinch of this price + = num_of_bedrooms * 1.0 # and a big pinch of that price + = sqft * 1.0 # maybe a handful of this price + = neighborhood * 1.0 # and finally, just a little extra salt for good measure price + = 1.0 return price

Step 2:

Bring each house into your function and check how far the estimate deviates from the correct price:

Use your program to predict house prices.

For example, the actual transaction price of the first property in the above table is $250000, and your function is valued at 178000, so you are 72000 short of this one.

Then sum the valuation of each property in your data set after deviating from the square. Assuming that there are 500 real estate transactions in the data set, the sum of the square deviations from the valuation amounts to $86123373. This reflects the "correct" degree of your function now.

Now, divide the total value by 500, and the valuation of each property deviates from the average. Call this average error the cost of your function.

If you can adjust the weight to make the cost zero, your function will be perfect. It means that, based on the data entered, your program values every property transaction exactly the same. And that's our goal-to try different weight values to keep the cost as low as possible.

Step 3:

Repeat step 2 over and over again to try all possible combinations of weight values. Which combination makes the cost closest to zero, it is what you want to use, as long as you find such a combination, the problem will be solved!

Thought disturbance time

It's too easy, isn't it? Think about what you just did. You take some data, enter it into three general simple steps, and finally you get a function that can evaluate the houses in your area. House price net, be careful!

But the following facts may disturb your mind:

1. Over the past 40 years, studies in many fields (such as linguistics / translatology) have shown that this general "stirring data soup" (the word I made up) learning algorithm has outperformed methods that require clear rules from real people. The "stupid" approach to machine learning finally defeated human experts.

two。 The function you end up writing is so stupid that it doesn't even know what "area" and "bedroom count" are. All it knows is to stir and change the numbers to get the right answer.

3. You probably don't even know why a special set of weights works. So you just write a function that you don't actually understand but can prove.

4. Imagine that there are no parameters like "area" and "number of bedrooms" in your program, but accept a set of numbers. Suppose each number represents a pixel in the picture captured by the camera installed on your roof, and then call the predicted output not "price" but "steering wheel rotation". So you get a program that can automatically manipulate your car!

It's crazy, isn't it?

What about "try every number" in step 3?

Well, of course you can't try all the possible weights to find the best combination. That will take a long time, because the number of attempts may be endless.

To avoid this, mathematicians have found many clever ways to find excellent weights quickly without having to try too much. Here is one of them:

First, write a simple equation to represent step 2 above:

This is your cost function.

This equation represents the extent to which our valuation program deviates from the current weight value.

If we graphically display all the possible weight values assigned to the number and area of bedrooms, we get a chart similar to the following:

Therefore, we only need to adjust the weight value so that we can "go downhill" to the lowest point on the graph. If small adjustments to weights keep us moving to the lowest point, then eventually we don't have to try too many weight values to get there.

If you remember a little bit of calculus, you may remember that if you take the derivative of a function, the result will tell you the slope of the function at any point. In other words, for a given point on the graph, it tells us which way is downhill. We can use this to head to the bottom.

So, if we take the partial derivative of the cost function with respect to each weight, then we can subtract that value from each weight. This will bring us closer to the bottom of the mountain. Keep doing this, and eventually we will reach the bottom and get the optimal value of the weight. Can't you read it? Don't worry, read on.

This method of finding the best weight is called batch gradient decline, and it is highly summarized above. If you want to understand the details, don't be afraid to move on.

When you use the machine learning algorithm library to solve practical problems, all of this is ready for you. But it's always useful to know some details.

Is there anything else you just skipped?

The three-step algorithm I described above is called multiple linear regression. Your estimation equation is looking for a straight line that can fit all the house price data points. Then you use this equation to estimate the price of a house you have never seen before based on where the house price may appear in your straight line. This idea is so powerful that it can be used to solve "practical" problems.

However, the method I show you may work in simple situations, and it won't work in all cases. One reason is that house prices do not always simply follow a continuous straight line.

Fortunately, however, there are many ways to deal with this situation. For nonlinear data, many other types of machine learning algorithms can be processed (such as neural networks or kernel vector machines). There are many ways to use linear regression to be more flexible, thinking of using more complex lines to fit. In all cases, the basic idea of finding the optimal weight value is still applicable.

Also, I ignored the concept of over-fitting. It's easy to come across a set of weights that perfectly predict house prices in your original data set, but not for any new house outside the original data set. There are also many solutions to this situation (such as regularization and the use of cross-validation datasets). Learning how to deal with this problem is very important for the smooth application of machine learning.

In other words, the basic concept is very simple, and it takes some skills and experience to use machine learning to get useful results. However, this is a skill that every developer can learn.

Is machine learning powerful?

Once you begin to understand that machine learning techniques can be easily used to solve seemingly difficult problems (such as handwriting recognition), you will have a feeling that as long as you have enough data, you can use machine learning to solve any problem. As long as you enter the data, you can see the computer magic trick to find the equation to fit the data.

But it's important to remember that machine learning can only be applied to problems that are actually solvable with the data you have.

For example, if you build a model to predict house prices based on the number of potted plants in each house, it will never succeed. There is no relationship between the number of potted plants in the house and the house price. So, no matter how it tries, the computer can't deduce the relationship between the two.

You can only model real relationships.

How to learn machine learning deeply

In my opinion, the biggest problem with machine learning at present is that it is mainly active in academia and business research organizations. For people outside the circle who want to have a general understanding rather than become experts, there are not many easy-to-understand learning materials. But the situation is improving every day.

Professor Andrew Ng's free machine learning course on Coursera is very good. I strongly recommend getting started from here. Anyone with a degree in computer science and can remember a little bit of math should be able to understand.

In addition, you can download and install SciKit-Learn and use it to experiment with thousands of machine learning algorithms. It is a python framework with a "black box" version of all standard algorithms.

That's all for what Machine Learning is. Thank you for your reading. If you want to know more about the industry, you can follow the website, the editor will output more high-quality practical articles for you!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.