How to realize algorithmic composition

You can use machine learning algorithm to compose music.

Traditional machine learning algorithms mainly refer to algorithms based on probability theory and statistics, among which Markov model is the most widely used.

Markov model mainly assumes that the state is Markov, that is, the probability of the next state depends only on the current state. A model mainly used is Hidden Markov Model (HMM), which has a layer of hidden variables and a layer of observed variables. As you can imagine, hidden variables drive observation variables to change, much like the relationship between chords and melodies.

At the same time, other Markov models are also used to form algorithms, such as Variable Order Markov Module (VMM) and Input-Output HMM(IOHMM). There are also many products that use this, such as MSR's SongSmith.

Theoretical development and application of algorithm

This algorithm was put forward by American R.O.Schmidt in 1979, which indicates that spatial spectrum estimation direction finding has entered a prosperous stage. It introduces the concept of "vector space" into the field of spatial spectrum estimation. After 30 years of development, it can be said that its theory is relatively mature.

Since 1980s, people have made extensive and in-depth research on the super-resolution spatial spectrum estimation algorithm based on feature decomposition, and put forward a series of efficient processing methods, among which the most classic is the multi-signal classification (MUSIC) algorithm, which needs one-dimensional search to find the direction of the source, while the computation of multi-dimensional search algorithms such as maximum likelihood (ML) and weighted subspace fitting (WSF) is greatly reduced.

The above contents refer to Baidu Encyclopedia-Algorithm Composition.