Calculus is undoubtedly useful but suffers from the Helen valley problem local gradients and so when a gradient descent function tell them the valley that might not be the global minimum so what you do is you’ve got to I’m sure they’re algorithms for this I’m just thinking about this that you you differentiate your classic differentiation but then you’ve got it you’ve got to do that several different times with different values or different spreads so you know not from one to two but then from 1 to 10 or 1 to 100 so kind of a log value to find out if that slope is consistent