Machine learning "1" (Python Machines Learning reading notes)

is still published as a reading note, not involving too many code and tools, as an understanding of the article to introduce machine learning.

The article is divided into two parts, machine learning Overview and Scikit-learn Brief Introduction, the two parts of close relationship, combined writing, so that the overall length, divided into 1, 22.

First, it's about machine learning. Key points are as follows:

1.1 Machine learning three main ways

1.2 Four processes

1.3 Python-related implementations (installation packages)

"1.1"

There are three main types of machine learning methods: Supervised learning (supervised learning), unsupervised learning (unsupervised learning), and reinforcement learning (enhanced learning)

Supervised learning (supervised learning):

The main purpose of supervised learning is to predict the future and classify it from existing data. First look at the basic definition:

The process of adjusting the classifier's parameters to meet the required performance, also known as supervised training or teacher learning, is adjusted using a set of known classes of samples.     Supervised learning is a machine learning task that infers a function from the tagged training data. In layman's interpretation, supervised learning is the analysis of a group (or groups of) known data, the optimal model of the condition, and the analysis of the data of the unknown result with this model, and the prediction and classification. This is widely used in our usual simple modeling and has a high degree of confidence. Thus, one of its major features is " there are training samples "The two most commonly used algorithms for          supervised learning are: Statistical classification classification and regression analysis regression      which algorithm to use, you need to choose according to the characteristics of the data. If the data is discrete, non-linear, statistically classified, and the data is linear, then regression analysis is chosen. The following two images illustrate the difference between the two:                           Figure 1   Statistical classification                                                                                                  Figure 2 Regression analysis          Reinforcement Learning (Enhanced learning)   &NBsp The purpose of reinforcement learning is to make the optimal decision in the corresponding environment. In essence, reinforcement learning has a result that we call a "return signal", which can be described as a "return function", which is similar to supervised learning. Therefore, this concept is first described in advance.     Unlike supervised learning, the return function of enhanced learning is not just a function value or a classification label, it is used to measure the merits of a step (or behavior), and the results are good. Therefore, the enhancement study is in the interaction, seeks the decision-making best. It is closely related to the environment.             The most typical example is the chess program. The program evaluates each step and gets the corresponding return function, and we only need to find a path with the largest return value (the sum of the returns per step), which is considered the best path.      unsupervised learning non-supervised learning     first-glance features: Dealing with unlabeled data or data of Unknown structure is mentioned earlier, the most important difference between unsupervised and supervised is whether the sample can be trained.     Two algorithms are: clustering and dimensionality reduction, from a very good understanding of the literal, clustering is the similarity of high objects converge into the same class. The data is not trained and analyzed directly. dimensionality reduction is to reduce the data dimension. The following two pictures, can be very good show:                                            3  Clustering                                          Figure 4 dimensionality reduction   "1.2" machine learning process         is mainly divided into four parts:  (1) preprocessing–getting data into shape      preprocessing is a critical step in machine learning and is one of the most fundamental steps in any data analysis. Real-life data is always not exactly in our format, and we cannot avoid the need to manipulate the data so that it becomes the format we need. Specific ways not detailed, you can refer to the previous articles, some mention, can also be used in practice, System learning   (2) Training and selecting a predictive model     In a nutshell, choose a suitable model.   (3) Evaluating models      Popular survey, error analysis and inspection   (4) predicting unseen data instances     If the model is ideal, you can apply     "1.3" Installing Python packages    numpy 1.9.1
scipy 0.14.0
Scikit-learn 0.15.2
Matplotlib 1.4.0
Pandas 0.15.2

Machine learning "1" (Python Machines Learning reading notes)