training -> loss -> (gradient + optimiser) -> parameters
eg cross_entropy
optimise parameters
We need an optimizer that can make use of the gradients to update the parameters. In complex cases, we might need more than one optimizer (e.g. GANs).
??
tensor = array
run jupyter on all ports:
./venv/bin/python3 venv/bin/jupyter-lab --ip=0.0.0.0