Chapter One Introduction
1.1 Basic concepts
Sample: A description of an event or object that abstracts something real into some mathematical form—a form that a computer can understand.
Feature engineering: A feature processing process that uses multi-dimensional features to describe samples, and can make features as simple or numerical as possible. In this process, the model established by using features can also be further optimized.
Sample space: The space where the feature vector of the sample is located, usually
expressed in fancy capitals.
Data set: Usually represented by a set, there are multiple samples in the set, the representation of the sample is a vector, and the number of features describing the sample in the vector is generally the same.
Model: In machine learning, data sets are generally divided into training and testing sets. By selecting a certain machine learning algorithm to train on the training set to obtain a model, and then use this model to test the test set to verify the effect of the model, you can adjust the parameters of the model, change the proportion of the training set and the test set, etc. method to optimize the model.
Label:
The label of the general-th sample is expressed mathematically as 
, so a complete sample is usually expressed as 
.
Classification: The value of the tag is discrete. It can be divided into binary classification and multi-classification. For binary classification, the positive class is usually recorded as 1, and the negative class is recorded as 0, ie
.
Regression: The value of the marker is continuous. The values may not be listed directly,
.
Generalization: The ability to make accurate judgments about the unknown based on the known. Algorithms and data are the key to the generalization ability: "Data determines the upper limit of the model, and the algorithm is to make the model infinitely approach the upper limit."
Data determines the upper limit of the model: the larger the amount of data and the more accumulated experience, the better the model effect; the more reasonable the feature numericalization, the more complete and detailed the feature collection, the better the model effect.
The algorithm is to make the model infinitely approach the upper limit: the effect of the model obtained by different algorithms is different, and the better the effect, the closer to the upper limit.
Distribution: That is, probability distribution. It is usually assumed that the sample space obeys an unknown "distribution"
, and the collected samples are independently sampled from the distribution, that is, "independent and identical distribution".
Hypothesis space: build model assumptions on data sets to predict future development trends.
Version space: A data set may have multiple hypothesis spaces, and different hypothesis spaces may learn models that can fit the training set. The collection of these models is the "version space".