This report describes the implementation of the data pre-processing for a novel anomaly detection technique. Proposed anomaly detection technique is based on using stochastic matrices as input for convolutional neural networks. Pre-processing step transforms raw data into 3d tensors combining stochastic matrices for a given event. Proposed solution for pre-processing is split into data reading part, which is parallelized with Dask and data processing part, which is parallelized with Spark. Data reading part runs on CERN’s Swan notebook and data processing part runs on a Hadoop server (specifically general purpose server Analytics).