1) Use .dt.total_seconds()
(df.purchase_time - df.signup_time).dt.total_seconds()
2) Or use np.timedelta64(1, 's')
(df.purchase_time - df.signup_time) / np.timedelta64(1, 's')
3) Or use .astype('timedelta64[s]')
(df.purchase_time - df.signup_time).astype('timedelta64[s]')
