Shapelets is a platform that hopes to help you as a data scientist. No, really. Time series analysis is becoming more and more important in the world of big data and data mining,, but without a time projection, it is impossible to analyze them. What our team realized was that the time series data management platforms on the market are based on architectures oriented to static and not dynamic problems. Trying to understand the behavior of a time series data using current platforms is similar to watching a movie frame by frame, one every few seconds, making it impossible to grasp its storyline.
Shapelets is the first state-of-the-art time series data management platform. The platform has an open source library of algorithms, and we will commercialize the differentials into which each data scientist will be able to integrate the algorithms of their choice.
The innovation of Shapelets is based on being a solution with an open source part, focused on the treatment and analysis of time series data, integrable with any type of storage platform, which automates machine learning functions through machine learning algorithms and where all algorithms are optimized for the available resources (CPU-GPU-Multicore).
What Are Time Series?
A time series is a collection of observations of a variable made sequentially over time, where the order of observation is important. The values of a time series are linked to instants of time (daily, monthly, yearly, etc.), so that the analysis of a series involves the joint handling of two variables; the variable under study itself and the time variable.
Time series predictions are an increasingly important field of Machine Learning that is often overlooked because many of the prediction problems are related to the time component.
Our algorithms make predictions about the future values of the time series. In general, the possible future values of a time series are calculated, taking into account all past observations equally. However, we focus on smaller time ranges to overcome the idea of "conceptual drift". That is, using only the last year of observations instead of all available data. When we deal with sets of time series, the problem is different: time series add an explicit order dependence between observations.
With the current rise of electronic devices (urban monitoring sensors, smart power meters, cell phones) and virtual agents (e.g., Twitter feeds) that detect, monitor, and track people and environmental activity, these types of data are continuously transmitted and interconnected. Two defining characteristics of these datasets, which are applicable to both IoT (internet of things) and social networks, are the temporal attributes of the time series and the correlation that exists between them. These datasets that present these temporal and graphical characteristics have not been adequately studied from the perspective of scalable Big Data management and analysis, despite the importance of analyzing this data.
What do Shapelets Algorithms Analyze?
Shapelets algorithms are the product of public-private collaboration, as a project of the Riverside University of California. Thanks to their work and this collaboration, we have groundbreaking algorithms that revolutionize time series analysis by automating much of the detection phase.
Have I seen This Before? Pattern Identification
This simple question is one of the key advantages and differences that Shapelets offers. With Shapelets algorithms, you will be able to identify patterns that you have seen before at a speed and precision unmatched by other tools in the market.
Unveil the power of pattern recognition with: A Summary of a Time Sequence
Another feature that Shapelets algorithms provide is the ability to summarize and classify periods of a sequence. This allows us to easily identify key information to respond to signal changes that may not be intuitive, eliminating the need for manual searches or summaries.
Unusual Patterns or Anomalies
Identifying and tracking unusual patterns has to be part of any time series analytics program. Automatic anomaly detection with Shapelets can save you time and help you check against past patterns.
Trend forecasting in various sectors can be done for the short, medium or long term, with different purposes in each case. In the case of the short term, the idea is to maximize the accuracy of the prediction, while in the medium and long term the overall trends become more important than a pinpoint accuracy.
Shapelets aims to revolutionize the time series analysis landscape with an open-source philosophy, integration with other platforms and above all empowerment of the data scientist. Find out more.