Stop searching after raising Skip Oversized model error
Opened this issue · 1 comments
Mojtaba1515 commented
Describe the bug
I have developed a code for searching different CNN models with a specific number of parameters, however after an architecture sampled during the optimization that exceeds the number of parameters, it stops searching by raising an error of "Oversized model: 180376 parameters. Skip model."
Expected behavior
It should skip that specific model and takes the second sample until it finishes the maximum number of trials.
TitoNicolaDrugman commented
Same here. I made a small work-around technique.
I am not familiar with Github, but basically this is what I made, it worked fine :)
I made a small variable outside get_model.
__run_once = True
This is the code, hope it will help!
__run_once = True # set this True, if the code suddenly stops after at least one tuning you need to edit it and set it to False
def get_model(args, hp):
global __run_once
trial_param1 = hp.Int("num_layers", 1, 30, step=1)
trial_param2 = hp.Choice("num_filters", [16, 32, 64])
trial_param3 = hp.Choice("dp_rate1", [0.0, 0.1, 0.2, 0.3, 0.4, 0.5])
trial_param4 = hp.Choice("dp_rate2", [0.0, 0.1, 0.2, 0.3, 0.4, 0.5])
trial_param5 = hp.Boolean("use_bias")
m = build_model_with_params(args, trial_param1, trial_param2, trial_param3, trial_param4, trial_param5)
if __run_once:
__run_once = False
return None
return m