Efficient Methods for Constructing Tensor Decompositions

In the 21st century, big data has already become an integral part of our lives. Almost all companies collect as much data as possible to analyse it in the future. In some cases, collecting this data is not as easy as it might seem. For this purpose, many companies use large exporting applications that collect data from various sources and put it into a large analytical database, where this data is used to predict revenues and analyse the company's performance over the previous year or quarter, for example. First, the data has to be collected from different sources, some is sent directly to the analytics system, while some need to go be extracted from partner companies' APIs. Each of these APIs works differently, but you need to standardise data uploading from them as much as possible to avoid code duplication. Secondly, the number of uploads increases with the company growth, so exporters should not operate for long periods. The whole system should be asynchronous. Thirdly, all possible errors should be monitored as much as possible, so a monitoring system should be added, all processes should be metered and logging should be thought through, but this should not slow down the export itself.