The original paper is on https://jcheminf.biomedcentral.com/articles/10.1186/s13321-018-0258-y

The following Binder link provides a readily installed Mordred and other tools to generate and process the calculated molecular descriptors, which can be used in downstream machine learning model building.

you could use a similar filtering code to filter those columns with values that is not a number

# Identify and drop columns with non-numeric values
non_numeric_columns = df.apply(pd.to_numeric, errors='coerce').isna().any()
df = df.loc[:, ~non_numeric_columns]