Towards increasing predictability of machine-learning research

<!--HTML--><p align="justify"> Dealing with system complexity is common problem for wide spectre of companies. It is always the case when people use machine learning for their needs. In Yandex as search engine company we are working with rapidly growing datasets, variety of data and different set of quality metrics. Usually research results in such environments are difficult to predict and tend to take unpredictable amount of time. </p> <p align="justify"> This talk describes several scenarios from Yandex everyday life that rely on machine learning techniques focusing on principles and instruments that help making research results more predictable both in terms of time required and quality of obtained results.</p>