First of all, thank you for making this library and helping others to perform HPO. I have a few doubts regarding this. As I am trying to perform HPO for one of my project where I am doing finetuning of ResNet50 on ImageNet-100 and want to train it for say 20 epochs and want to optimize hyperparameters like lr, step_lr, step_lr_gamma, etc. So here I am a bit confused about the min_fidility, max_fidility, total_cost, and cost. So currently I have taken these values as -
min_fidility = 2
max_fidility = 20
total_cost = 70000
cost = end_time-start_time
I am bit confused if this will give it enough configurations to evaluate the best set of hyperparameters.
First of all, thank you for making this library and helping others to perform HPO. I have a few doubts regarding this. As I am trying to perform HPO for one of my project where I am doing finetuning of ResNet50 on ImageNet-100 and want to train it for say 20 epochs and want to optimize hyperparameters like lr, step_lr, step_lr_gamma, etc. So here I am a bit confused about the min_fidility, max_fidility, total_cost, and cost. So currently I have taken these values as -
min_fidility = 2
max_fidility = 20
total_cost = 70000
cost = end_time-start_time
I am bit confused if this will give it enough configurations to evaluate the best set of hyperparameters.