M4ML C2_W2 - learning rate question

I have a question about the learning rate.
Maybe this is a naive question, but if we are able to draw the graph, why don’t we use computer vision to identify the lowest point in the graph?
Or is drawing the graph the actual problem?

Obtaining the graph performance of the network as the learning rate changes is the hardest thing because you have to run and test the model as learning rate values change.

Finding a minima once you have the graph is easy.