What is machine learning

Learning two words is a human behavior. If a machine also wants to learn, it will naturally have to learn from us

How do we learn?

Of our people, all the time to collect data from the surrounding environment, eyes, ears of breath these sense organs, skin is formed among the data in our mind about things cognitive experience, again after a summary of the brain, are formed for a class of things the understanding of the law, when we meet with the new things, we are all according to the law of experience and cognition in has been formed to understand

Therefore, machine learning is a category of bionics, and it is natural to simulate the process of human cognitive affairs, thus forming the core logic of machine learning:

The cognitive model can be automatically analyzed from historical data and used to predict unknown data

Machine learning is essentially this line of words, taking features from data, and putting them together in a cognitive model, or pattern, that allows you to learn like a human

Here, for example, is the comparison between machine learning and humans, which are the same thing

For example, here comes a dog you’ve never seen before

  • How do people know? When people saw a dog in the past, they automatically recorded some characteristics of the dog: body shape, face shape, nose shape, color, bark, behavior and other characteristics. According to these characteristics of the seat, naturally know this is a wang, a later inquiry, the original dog is known far and wide to dismantle small expert: husky
  • How does the machine know? As a scholar, we extract significant feature points from historical data, value and record them, and form a model. When new data comes in, we will naturally know that it is a dog based on the comparison of its characteristic values

Machine learning is about analyzing feature points from pictures of cats and dogs to form models of cat and dog cognition

Some people say machine learning can predict. What’s going on?

In fact, the same as above, from the historical data to find the relationship between the flip characteristics (location, type, size, floor, property) and value (here is a set of formulas), and then apply this model (formula) to calculate the price of other housing data. Actually, it feels like doing math, so there’s a lot of data thinking going on in machine learning

Ok, this is machine learning, don’t think this concept is too complicated, we are all doing development, do you know how to make the complex simple, difficult according to simple understanding is the right way

The data set

Data and algorithms are the core of machine learning, not just any kind of data can be machine learning. Machine learning has its own requirements for data, called data sets in machine learning

Let’s look at machine learning as a definition of data:

Data set: eigenvalue + target value

Typical data sets:

  • The size of the house, the location, the floors, the orientation, these are the characteristics of the house
  • The price of the house is the target value
  • Each row of data is called:sample

Why should there be a target value, for example, if we want to predict the price of housing, in fact, is to find a final value, so there should be a target value

Datasets can have no target values:

So machine learning data set characteristics: eigenvalue + target value, some data can be no target value

Machine learning algorithm classification

The purpose of machine learning is to solve problems by using historical data to train models to predict future prices, trend changes and types of ownership. These are all specific problems to deal with. The target problems in machine learning are divided in detail, and different types of problems are supported by different algorithms:

Notice that the classification of the problem is based on the target value, think about it, what is the target value, isn’t it for what, for example, if the target value is a value, then the goal of machine learning is to predict the value; The target value is the category, and the goal of machine learning is to predict the category, so say:

  • Target value: Category –Classification problem
  • Target value: continuous data –Regression problems
  • Target value: No –Clustering problems, unsupervised learning

Let’s practice and see what the problems are:

  • Predict what the weather will be tomorrow– Regression problem
  • Predict whether it will be sunny or rainy tomorrow– Classification problem
  • Face age prediction– Classification or regression
  • Face recognition– Classification problem

Different problems are supported by different algorithms:

  • Supervised learning:
    • Classification:K-nearest neighbor algorithm, Bayesian algorithm, decision tree and random forest, logistic regression
    • Regression:Linear regression, ridge regression
  • Unsupervised learning:
    • Cluster: k-means

Let’s look at this first, and know what kind of problem the data of the target value corresponds to, and then deepen our understanding…

Machine learning development process

Any project has its own conventional process, but machine learning has the same, as follows:

  • To get the data
  • The data processing
  • Characteristics of the engineering
  • Machine learning algorithm training - model
  • Model to evaluate
  • application

The data processing part is actually processing the default values of the data and so on. Feature in feature engineering is the feature value in machine learning data. Feature engineering will process the feature value and target value into the most acceptable and suitable data for the following algorithm according to the type of deep learning algorithm adopted later. In many cases, feature engineering is the key and difficult point

OK, we first remember so much, I will make up the back, I also just began to see ~

Library and resources for machine learning

1. First of all, the algorithm in machine learning is very difficult, so difficult that ordinary people can’t do it. All of them are professors and research institutions from internationally renowned universities

Most of the algorithm design of complex models is done by algorithm engineers, and we do:

  • Analyzing a lot of data
  • Analyze specific business
  • Apply common algorithms
  • Feature engineering, parameter tuning, optimization

80% of the work is in the processing of data, data cleaning, data processing, characteristic engineering and so on, in the enterprise are generally processing data, algorithms are mature programs, few of their own research, unless it is a large company with ambition and money, domestic ali bar

2. How to learn do you know how to learn in university

Let us ordinary people or step by step, the biggest domestic development in recent years, is to become cabbage. The python+ Machine learning course is available for 20 quick dollars on the X package, which includes detailed python basics. If you just want to watch the machine learning part, there are related videos on the B website:

  • Ai & Machine learning Quickstart 1/3
  • Ai & Machine Learning Quickstart 2/3
  • Ai & Machine Learning Quickstart 3/3
  • Dark horse B station account under the video, also part

What is the level of entry

  • Learning to analyze problems, the purpose of using an algorithm in machine learning, what tasks do you want to have an algorithm to accomplish
  • Master the basic idea of algorithm, learn to use different algorithms according to different problems
  • Learn to leverage open source libraries and frameworks to solve real-world problems

3. Books:

Python is the open source library and framework for machine learning. Ha ha ha… These are the main libraries

  • The above isSKlearnThis library is a machine learning library, the others are deep learning libraries
  • TFThis is the hottest. Google produces the must-have
  • pytorchandcaffe2It’s all facebook
  • theanoisTFThe predecessor of
  • chanerispytorchThe predecessor of

The so-called use of new not old, everyone knows, well, so much information

Examples of machine learning

Machine learning is a huge departure from traditional programming, where we learn how to use off-the-shelf apis and code organization to output a lot of code, with W lines of code, right

However, machine learning is different. Machine learning is more like learning theories and tools in machine learning to solve practical problems. The output is limited to several hundred lines of code or even less

To make it clear, HERE are some classic examples:

  • Predict Titanic survival rate – 10 minutes