Supervised learning (supervised learning): The reason to call supervised learning is because we tell the algorithm what we want to predict. The so-called supervision, in fact, is whether our intentions can directly influence the forecast results. Typical representatives: Classification (classification) and regression (regression).
Unsupervised learning (unsupervised learning): In unsupervised learning data, there is no label (label, for category distinction, etc.) and target value (target values, for regression predictions). In general, if we want to group data items that have similarities, this behavior is "clustering" (clustering). In addition, if we want to know some of the probability values of data, then this behavior is called "Density estimation" (density estimation). Finally, unsupervised learning may also be used to reduce the dimensionality of multi-feature (feature) data, eliminating some unimportant features, allowing us to observe data in low-dimensional space.
Steps to develop a machine learning application:
1. Collection of data;
2. Prepare input data;
3. Analysis of input data;
4. Input data detection, or sample preprocessing (reject bad data);
5. Training sample, get model;
6. The testing model is based on the accuracy and accuracy of the prediction;
7. Use the model for practical applications.
The world of machine learning is a world of probabilistic statistics, and samples can be transformed in different spaces to highlight certain features and simplify the description of rules.
Machine learning Mlia Notes (i)