Encoding Time Series as Images. Gramian Angular Field Imaging | by Louis de Vitry | Analytics Vidhya | Medium

Analytics Vidhya is a community of Generative AI and Data Science professionals. We are building the next-gen data science ecosystem https://www.analyticsvidhya.com Follow publication Top highlight 813 7 Listen Share The Deep Learning boom is largely fueled by its success in computer vision and speech recognition. However, when it comes to time series, building predictive models can be gruesome (Recurrent Neural Networks are difficult to train, research is less applicable, and no pre-trained models exist, 1D-CNN can be inconvenient). To leverage the techniques and the insights brought by the recent developments of computer vision, I will present and discuss one way to encode time series as images: the Gramian Angular Field. Beside the math pre-requisites (min-max scaler, dot-product and Gram Matrix), this post will contain explanations and solutions on: as well as Python gists of: TL:DR: We perform a polar encoding of the data followed by a Gram Matrix like operation on the resulting a

Analytics Vidhya is a community of Generative AI and Data Science professionals. We are building the next-gen data science ecosystem https://www.analyticsvidhya.com Follow publication Top highlight 813 7 Listen Share The Deep Learning boom is largely fueled by its success in computer vision and speech recognition. However, when it comes to time series, building predictive models can be gruesome (Recurrent Neural Networks are difficult to train, research is less applicable, and no pre-trained models exist, 1D-CNN can be inconvenient). To leverage the techniques and the insights brought by the r