Start tuning with default values

[lcrmorin]

After spending some time handtuning my NN I came to use keras tuner. When setting the tuner up I specified the default values as the values that worked better when hand tuning. I expected the tuner to start with those values in the first round.

I'd like a first round with default value when they are specified, or an option to do so.

It's not exactly clear what whould the alternative be. Feed the tuner with my handtuned values and loss ?

[haifeng-jin]

It is because this function is building the model first for several times. If you can override it to do nothing. The searcher would start with the default value. https://github.com/keras-team/keras-tuner/blob/master/keras_tuner/engine/base_tuner.py#L111

[Erik-BM]

So you suggest this as an alternative:

class DefaultTuner(kt.BaseTuner):
    def _populate_initial_space(self):
        pass

and set trials=1 to get default parameters?