Automl algorithm for working with time series


Automl algorithm for working with time series


In companies, only a small number of the most experienced managers have full knowledge of a specific area of the organisation’s functioning, including information about:

  • products and services, in particular replenishment amounts, ways to increase sales in the thick of fierce competition and the possibility of decreasing demand;
  • employees, including the number of people performing the production function;
  • company assets, e.g., daily stock quotes, etc.

As a result, the company is highly dependent on these managers, who may not be able to process the full amount of data, make mistakes in forecasting and, therefore, miss out on many opportunities.


An AutoML algorithm that identifies consistent models and patterns based on historical data as well as time series information and associated numeric/category traits.

As a result, it is possible to forecast the value of target variables with an estimate of the confidence interval over a given horizon.

For example, the algorithm makes it possible to better identify sales opportunities by drawing managers’ attention to certain events, while at the same time reducing the burden on the most experienced employees

Technology stack

  • Python
  • Pytorch
  • Sklearn
  • XGBoost