This article is an introduction to Machine Learning (Machine Learning). You will learn the basic tasks, types, and functions of training on models, familiarize yourself with some algorithms (algorithms) and functions of modeling (modeling). The article will be able to give answers to many other questions.
ML (machine learning) solves many problems today, making life easier for many. It is enough to put in the PC an algorithm (algorithm) for finding solutions in order to be able to comprehensively use statistical data and display various patterns, and make predictions.
Machine learning consulting technology appeared in the middle of the last century when the first checkers’ programs began to appear. The essence has changed little over the years. But the computing power of our computers has changed, as a result, patterns and forecasts have become more complex, the number of tasks and problems to be solved has increased.
To start the training process, you need to load the initial data, called a dataset. Use tagged animal images if you like. It is on them that the algorithm will learn how to process requests. When the training is over, the model will be able to independently recognize the necessary images, and the tags are no longer needed. Learning will not stop there at all but will continue as you use the program. The more data is eventually analyzed, the darker the recognition will be.
Today ML recognizes drawings, faces, landscapes, numbers, letters, and objects. The same machine learning grammar checker feature is found in almost every text editor. And this function takes into account both spelling and other points, including lexical combinations, jargon, and other subtleties of the “great and mighty”. There are also programs that generate news texts themselves – even human participation is not required (copywriters, think about it!)
Machine learning tasks
The tasks to be solved can be classified into several categories:
- regression – we are talking about a forecast based on a sample of objects with different characteristics;
- classification – here we get a specific answer based on a set of features (“yes” or “no”);
- clustering – this term refers to the distribution of data into certain groups (for example, the level of the client’s solvency);
- reduction in dimension – a large number of features are reduced to fewer so that it is more convenient to visualize them later;
- detection of anomalies – anomalies are separated from the usual (standard) cases. The anomaly may coincide with the application, but only at first glance. A real example is bank card fraud.
Separation
Depending on the size of the dataset, in some cases, you may only need a small portion of it. This is commonly referred to as sampling. From the selected part, the data should be divided into two groups: one for use by the algorithm, and the other for evaluating its actions.
Education
This stage is actually aimed at finding a mathematical function that will accurately perform the specified task. The training varies depending on the type of model used. Drawing lines in a simple linear model is a learning curve; generating a decision tree for a random forest algorithm is also learning. Changing the answers when building a decision tree will help you correct the algorithm.
To make it easier, let’s focus on neural networks.
The bottom line is that the algorithm uses a piece of data, processes it, measures the efficiency of the processing, and automatically adjusts its parameters (also called the backpropagation method) until it can consistently produce the desired result with sufficient reliability.
Grade
After an algorithm has performed well on training data, its performance is evaluated on data that it has not yet encountered. Additional adjustments are made if necessary. This process helps prevent overfitting, a phenomenon in which the algorithm works well only on training data.
