The paper explains various methods and techniques of time series modeling and then the presentation of predictive ones models and results achieved by each of them. During the introduction and introduction to the time series it is important to note the things to keep in mind when modeling. Time series require good understanding the data we observe and decomposing the parts that make it up. Time series are specific because it is mostly a single observed variable, which is in the prism of a certain time intervals. In addition, it requires an understanding of the data over time, and good detection of phenomena such as seasonality and trend. Procedures from the very beginning of model construction such as data cleaning, testing are presented stationary, visualization and the like. We went through exponential smoothing methods, ARIMA models and TensorFlow package Structural Time Series. The paper shows how each of these methods can accomplish strongly good results. TensorFlow STS, however, differs from other methods since it is a new library which combines traditional statistical methods that are linked to time series and various algorithms and principles of machine learning. It also takes as a perspective Bayes' thinking of updating the evidence with with the arrival of new information.