fmin_xx( func, x0, fprime=None, epsilon=.001 ... )

fmin_tnc( func, x0, approx_grad=True, epsilon=.001 ... )  or
fmin_tnc( func_and_grad, x0 ... )  # func, your own estimated gradient