Hi! Is this trained on both malicious and benign samples from the EMBER set? Unclear from the documentation whether the weak supervision of importance labels comes from exclusively malicious distributions. Using this tool in a research setting where this is a potentially important distinction.
Thanks!