Choosing a persistent storage for data mining task
Paweł Kasprowski · 2013
The amount of data available for mining or machine learning is increasing. Therefore one of the main problems of nowadays mining is decision how to persistently store that data in the way that it is easy and fast to load and save by mining algorithms. When data is too big to fit in the memory, there are two common ways to handle it: text or binary file in own format or ready-to-use universal database engine. Both have advantages and disadvantages. As for database engine, the most popular storage is a relational database server. Recently another promising option became non-relational databases like document-oriented databases. The work presented in this paper analyses how different storages behave for big amounts of data. Experiments compare efficiency of these storages for some classic mining tasks