TABot – A Distributed Deep Learning Framework for Classifying Price Chart Images

In this work, we propose a framework called TABot, a distributed deep learning (DL) framework for classifying price chart images. The deep learning engine we outline consists of an ensemble of convolutional neural networks employed for classifying price chart images. We present three workflows that compose our framework. The first is the data sourcing workflow: a distributed asynchronous pipeline to collect price data and programmatically generate candlestick charts. We measure the processing times of our distributed solution relative to a synchronous analog to identify a nominal processing time differential between the two solutions. The second is the model training workflow. We again leverage a distributed asynchronous pipeline to train each convolutional neural network in a parallel fashion. We measure the processing time of our parallel solution and compare to a synchronous analog. The third is the prediction workflow. We introduce a simple scheme for collecting the prediction output of each component model in an ensemble network model. Our results support the viability of convolutional neural networks to classify price chart images. The TAbot architecture additionally highlights the benefit of utilizing elastic computing environments to manage computational and data persistence costs incurred by deep learning frameworks.