The project was to build and release the first publicly available code evidence dataset called MDACE on a subset of the MIMIC-III clinical records. We believe that the release of MDACE will greatly improve the understanding and application of deep learning technologies for medical coding, and benefit machine learning models for classification.