Suggest a better method to provide the training data and its labels to model.fit()

[itsSrijan]

I have two variables called timesteps and nosamples which I am hyper tuning apart from others. And my input shape to the model depends on these two values. Hence, for every new pair of values of these, I have to change the shape of my training data and its label. To overcome this issue, I get the training data and its label inside build_model() and return them along with the model.

def build_model(hp):
  timesteps = hp.Choice('timesteps', [2, 4, 8, 16], ordered = False)
  nosamples = hp.Choice('nosample', [ 32, 64, 128], ordered = False)
  .......
  train, train_label, _, _ = train_test_splitting(DFS, UFS)

  model = keras.Sequential()
  model.add(InputLayer(input_shape = (timesteps, nosamples)))
........
  return model, train, train_label

I am providing train and train_label to model.fit() this way

class MyTuner(RandomSearch):
  def run_trial(self, trial, *args, **kwargs):
    hp = trial.hyperparameters
    model, train, train_label = self.hypermodel.build(hp)
    kwargs['batch_size'] = hp.Choice('batch_size', [16, 32, 64, 128])
    return self.hypermodel.fit(hp, model, train, train_label, *args, **kwargs)

I know the best way to tune batch_size is this, but I couldn't understand how to use it for train and train_label.

The method I employed is working fine but I think there is a more efficient way to do the same that I don't know. Any help would be welcome.

Thanks for the patience to read till here.

[haifeng-jin]

You can access the training data in HyperModel.fit().
It should be something like this:

class MyHyperModel(HyperModel):
  def build(self, hp):
    timesteps = hp.Choice('timesteps', [2, 4, 8, 16], ordered = False)
    nosamples = hp.Choice('nosample', [ 32, 64, 128], ordered = False)
    ...
    return model
  def fit(self, hp, model, DFS, UFS):
    timesteps = hp.get('timesteps')
    nosamples = hp.get('nosample')
    strain, train_label, _, _ = train_test_splitting(DFS, UFS)
    ...

...
tuner.search(DFS, UFS)