Introduction

--Machine learning --Application and science of algorithms to understand the meaning of data --Exciting fields in computer science ――This chapter deals with the main concepts of machine learning and their types. --Contents to be handled --General concept --Three types of learning and basic terms --Components for system design --Python setup --Sample code - python-machine-learning-book/code/ch01/ch01.ipynb ――In the following summary, the code and formula are not described. I'm sorry.

1.1 "Intelligent machines" that turn data into knowledge

--Large amount of data --Structured data --Unstructured data --Examples of application in daily life --Email spam filter --Character / voice recognition software

Search engine --Competition with Go players

1.2 3 types of machine learning

--Supervised learning --Unsupervised learning --Reinforcement learning

1.3 Future prediction by "supervised learning"

Target --Learning a model from training data so that you can predict unknown and future data --Supervised data --A set of samples for which the desired output signal is already known --If there is an email spam filter, is it "spam" or "not spam"? --Example --Classification: Has a discrete value class label --Regression: Output signal has continuous value

1.3.1 Classification for predicting class labels

Purpose --Predict class labels for new instances based on past observations --Class labels are discrete and out-of-order values (affiliation) --Binary classification --Email spam filter --Multi-class classification --Handwritten character recognition

1.3.2 Regression for predicting continuous values

Purpose --Given multiple predictors and continuous response variables, explore the relationship between those variables so that you can predict the outcome. --Linear regression

1.4 Reinforcement learning to solve dialogue problems

Target --Developing a system (agent) that improves performance based on interaction with the environment --Can be regarded as a field related to supervised learning --Information about the current state of the environment also includes reward signals ――This feedback is not a label or value of the correct answer, but quantifies the performance of the action measured by the "reward" function. --Maximize rewards --Trial and error approach --Do not use model --Cram school planning --Use the model --Example --Chess engine --Reward wins or loses

1.5 Discovering hidden structures through "unsupervised learning"

--Unsupervised learning --Handling unlabeled data or data of unknown structure

1.5.1 Discovery of groups by clustering

--Clustering (unsupervised classification) --Exploratory data analysis that can structure a large amount of information as a meaningful group --Exploratory data analysis: Calculating data statistics and visualizing the distribution to exploratoryly derive knowledge about the data. --Example --Discovery of customer groups in marketing

1.5.2 Dimensionality reduction for data compression

-(Unsupervised) Dimensionality reduction --Compress data into lower dimensional subspaces while preserving most of the relevant information

Purpose --I want to avoid processing a large number of values because the storage space and computing performance are limited. --Data visualization --Method example --Unsupervised dimensionality reduction --Principal component analysis --Kernel principal component analysis --Supervised dimensionality reduction --Discriminant analysis

1.6 Basic terms and notation

sample
Feature value --Target
linear algebra
vector
queue

1.7 Roadmap for building a machine learning system

--General workflow when using machine learning for predictive modeling

Pretreatment
Learning
Evaluation
Forecast

1.8 Preprocessing: Data shaping

Preprocessing --Convert to the format required to optimize the performance of machine learning algorithms --Make the selected features the same scale --Convert features to a range of 0 and 1 --Converted to a standard normal distribution with mean 0 and variance 1. --Depending on the extracted features, there is a high degree of correlation and constant duplication. --Dimensionality reduction --Check if it can be applied (generalized) to new data sets --Split the dataset into a training dataset and a test dataset

1.8.1 Predictive model training and selection

--Comparison of several algorithms is essential to train and select a good model --Indicator for measuring performance --Correct answer rate --Estimation of model generalization performance --Split training dataset for training and validation, cross-validation --Hyperparameter optimization

1.8.2 Model evaluation and unknown instance prediction

--Evaluation of generalization error --Apply the model to the test dataset and check how well it will perform against unknown data --The parameters of the above procedure such as feature scaling and dimensionality reduction are retrieved only in the training dataset.

Python Machine Learning Programming Chapter 1 Gives Computers the Ability to Learn from Data Summary