938e9716faa50db5695d937cae0b30d2cdbfd07c,conceptnet5/vectors/transforms.py,,l2_normalize_rows,#,67

Before Change

    Pandas-approved way to represent missing data, so Pandas should be able to
    deal with those.
    
    row_norms = np.sqrt(np.sum(np.power(frame, 2), axis="columns")) + offset
    return frame.div(row_norms, axis="rows")


def subtract_mean_vector(frame):
    

After Change

    distance are all 1. This enables cosine similarities to be computed as
    dot-products between these rows.
    
    index = frame.index
    return pd.DataFrame(data=normalize(frame, norm="l2", copy=False, axis=1), index=index)


def subtract_mean_vector(frame):
    

Instances