Daily Dose of Data Science – Day 4 – Automate the time-series forecasting using Auto_TimeSeries
Let’s get you yet another Daily Dose of Data Science! This time we will discuss on an important topic about AutoML for time-series forecasting as time-series is one of the most common type of datasets any data scientist might come across and time-series data may require good amount of tuning to get generalized result. So, we will discuss about the python framework of Auto_TimeSeries or Auto_ts which can help us to achieve automated time series forecasting and hyper-parameter tuning.
Auto_TimeSeries is one of the most widely used automated time series forecasting framework which enables you to build multiple time series models with just single line of code. It is a complex model building framework for time series data and effectively automates many complex hyper-parameter tuning tasks. The default settings are easily configurable as well. Auto_ts can rapidly build predictive models based on Statsmodels ARIMA, Seasonal ARIMA and Scikit-Learn ML, and even deep learning based FB Prophet model. After running all the popular time series models, it has the ability to automatically select the best model which gives best score specified like all standard AutoML frameworks. Please go through the official GitHub project repository of this framework to learn more: https://github.com/AutoViML/Auto_TS.
The setup
Like all other standard python packages, the setup for this framework is also very easy and can be done using the pip installer in any OS environment.
pip install auto-tsDealing with large datasets
Even when working with time-series data, dealing with large datasets in Pandas can be very challenging. So, for any version above 0.0.35 for this framework allows you to load your file into Dask dataframes by just providing the name of the file and if it is too large to fit into a pandas dataframe, Auto_TS will automatically detect and load it into a Dask dataframe.
Both Univariate and Multi-variate Time series forecasting
Time series forecasting can be based on a single time-series variable (known as Univariate Forecasting) and as well as for multiple variables (known as Multi-Variate forecasting). This framework supports both these types and can be seamlessly used for both Univariate and Multi-variate analysis. Example code notebooks can be found at: Univariate Example, Multi-variate Example.
I will highly recommend all data scientist who deal with time series data alot in their daily lives to explore this wonderful framework. Kudos to Ram Seshadri for developing the framework and Nikhil Gupta for significantly expanding in functionalities and features of this framework. Visit https://github.com/AutoViML/Auto_TS to find out more! Stay tuned for another daily dose of data science and please feel free to like, share, comment and subscribe to my posts if you find it helpful!