About Machine Learning ( just a introducation :-)
well,so I finally demind to use English to complete my blog whose context is related to Machine learing. Because my English is so poor and I want to get a better English skill. On the other hand, in order to emhance my understudying of machine and deep learning,I will make a effect to write every blog which is about AI in English. If this decision causes any confuse and discomfort ,please give me a feedback in the comments.
What is Machine learning?
so,in a word, ''make Machine to learn how to solve a problem".
In the past , people used to solve diffcult problems by progammed code. We should collect dataset , find same data point and group them, design a algorithms,analysis cost in time or space.........in result, computer only needs to translate high language to machine language and count a result , while we were studying hardly.
In Machine learing, the roles is revolved. Making computer solve the qusetion without programmed codes, is the target of us.
it is very abstract,I know,becasue ml`s basic is.......
mathematic
Main Machine learning algorithms
In the first blog , I dont`t want to explain what is ai and how to start machine learning,even through many of you like me, need a paper to fill you science experiences.I understand you.
but, before the first step,let's learn two kinds of the widest being used and rapidest advancement algorithms:
- supervise learning
- unsupervise learning
supervise learning
definity:computer learns by being given "right answers",then outputs a target "y" with target "x" that computer never sees it.
Simply , supervise means use the dataset to predict 'y' with a new variable "x" .
Maybe it is shill vary abstract.Now I will show you two supervise learning algorithms: regression(回归) and classification(分类)
REGRESSION
Let's set a scene:
you are a real estate agent（房屋中介）.you customer , MaLeJi want to sell his real estate in NingBo and HangZhou. He comes here and consult you.
you have some records:
This table shows that house has different price with different size in feets. But MaLeJi's house size is not match any records in this table.
In this situation，you should recall “regression”，and use it to fit a model. This model will predict the price with a new 'input'.
For example ,you can fit a linear function,like this
Input you new data, after the training from early dataset.you will get a predict value "y"
this is a example case of regression.by the way,it is a linear model.Exceptint it,we can use other model to fit it ( such as curve )
CLASSIFICATION
When I read a book called 《算法图解》, I get a simple example.
you are a storer(sell fruit), you have a abundant experience of recognize apple and tomato.
It is simple to divide them by size. you know the size range of apple and tomato.
OK,in this process,you built a classification model. when a fruit is passed in you hand,you can judge it quickly. look like this:
Unlike the regression, classifiaction only get small number of outputs.Like this example,we put new point in X-axis.Maybe the point has a closer distance to point group repesenting apple , then this item is a apple.
What‘s more , the dimension standards of classifition is not unique.
unsupervise learning
unsupervise means that computer only need input,not output 'y'.Algorithm has to find the structure in dataset.
In a word : no output , just to find data structure.
let's see a example: clustering
as a front-end programmer，I always have to face to the browers search engine(by the way ,I hate it......).In a searching , you will see some similar topics while searching a topic.
such as this: if you search 'panda' and click
there are many relative topic.
clustering is a kind of unsuperivse learning.computer find similar attirtudes from dataset,and group them which has similar attritude to same cluster.