Explain machine learning through the little things around you.

explaining machine learning through the little things around me

A very good example of a person who does not know what machine learning is, and how to do it.

Today, I saw a question from the Quora: How to give students who are not CS, to students who do not know machine learning and data mining, to understand what is machine learning and data mining.

One of the answers is very good, take the example of a mango to explain to you. Teachers should also use a number of similar examples to inspire students ' thinking.

===

Buy Mango

One day, you want to eat mango, you go to the side of the stalls to buy mangoes, stalls have a lot of mango ah, you can pick the mango by hand. After jumping, the stall owner will give you weigh, how many pounds? According to the weight to pay the money.

If you buy mangoes, you can choose the sweetest, ripe mango as long as it is not a heavy taste or a unique taste. Because you pay according to the weight, not according to the degree of sweetness or ripe, although the stall owners sometimes pick a good mango out of a pile of price alone, but this stall owner did not do so.

Your grandmother once told you that buy Mango to buy gold, yellow orange orange amber hot, such the sweetest, do not buy those light yellow, those are not ripe.

OK, so you have a little experience, although this experience is what others directly teach you: Buy mangoes, buy gold. You in the stall, picked some golden, weighed pay, go home. So this is the end of the story? Don't worry, the following.

life's not that simple .

You go home and eat mango happily, but you find that not every mango is so sweet and partly not sweet. Alas, the grandmother's experience is still insufficient ah, although eat more salt than I walk, but only through the color of the mango sweet not sweet, not very reliable.

What kind of mango sweet do you recall? It seems to be big and golden sweet, and those little golden Mango, half is not sweet. (bought 100 golden Mango, 50 large, all sweet, and 50 small, 25 of which are not sweet.) ）

Well, you finally summed up a rule of thumb: Big Golden is the sweetest, haha. You're happily shopping for mangoes. Shit, you are familiar with, you trust that the stall owner has gone. So you have to change a stall to buy mango, but the new owner of the Mango is produced from different places, you can not summarize the experience may not, you do not know if you can migrate the past (transfer learning), so you start again to try it, found here small, light yellow is the sweetest!

One day, your cousin came to you to play, want to eat mango, but she does not care sweet not sweet, she likes to eat juicy. Alas, the experience of the past has not worked. You can only perform a new round of experiments with the goal of a juicy mango (the optimization target has changed). You also conclude that the more soft the more juice.

You go abroad to read PhD, the mango here and your hometown is not too bad, here the best green to eat. PhD After graduation, you are married, the wife does not like to eat mango, like to eat apples. You accumulate a rich selection of mango experience rules are not, perhaps some of you can transfer the past. You have to start all over again with a round of experiments to see how some of the apple features are related to the delicious and tasty. Although the process is boring, but you do, because you love her.

List of rules

You want to make a choice of how to choose the Mango (apple) program, so that on the computer, even with your mobile phone scan, you can automatically pick out a lot of delicious mango. Because you have accumulated some rules, you can achieve this:

if (color is bright yellow and size are big and sold by favorite vendor): Mango is sweet.
if (soft): Mango is juicy.
etc.

But you want to ah, these rules more and more, the combination of features between the more and more trouble, management, use is very troublesome. Including write program implementation Ah, who would be stupid enough to write so many if then.

Machine Learning

Machine learning algorithms are the evolution of common algorithms, smarter and more automatic. See how to define the mango problem as a standard machine learning problem.

Randomly selected a mango on the market as a target to be studied (training data). You can use a table to describe the relationship between Mango properties and types, and each row can contain a mango's data, including the Mango's physical properties (feature): color, size, shape, hardness, origin, and so on, as well as the type of mango (output variables ): Sweet, ripe, juicy. Then this is a multi-classification problem, or regression problem, automatically from the data to learn the characteristics of the relationship between the Mango type and so on.

If you use a decision tree algorithm, the model looks like your rule base, and of course you can use other models, such as linear models, which are linear combinations of features.

Next time you go to the market, collect a mango of each indicator characteristics, throw into your model, the model tells you what type of mango? Is it cooked? Juicy, huh?

various methods

Even your choice of mango model, a little change can choose apples, migration learning.

Even your model will become more and more better and learn incrementally as new samples and new mango species come in.

。。。。。

Do you know anything about machine learning?

Explain machine learning through the little things around you.