I have a few models that I've been hyper tuning with Keras Tuner BayesianOptimization, doing 100 trials at 4 epochs each, and not sure if it's just coincidence or what, but the very first trial has been the best for several of these models.
I can't help but think that I'd be better off just doing a random search or hyperband.
I'm also just figuring this stuff out, so I'll admit I'm a bit naive here.
