[SOLVED] Positional arguments for HyperOpt fmin() function

Positional arguments for HyperOpt fmin() function

I am trying to user hyperopt.fmin to optimize the hyperparameters of a XGBoost classifier. I have an objection function:

from hyperopt import fmin

def objective(space, x1, x2):
   'do whatever to define loss_array'
    return {'loss': -loss_array.mean(), 'loss_variance': np.var(loss_array, ddof=1),'status': STATUS_OK, 'scores':score_dataframe}

I'd like to have the optional argument x1 and x2 (for e.g., if I want to specify a different loss function).

Then, I can minimize:

argmin = fmin(
  fn=objective,
  space=space,
  algo=tpe.suggest,
  max_evals = 700,
  trials=trials,
  rstate = np.random.RandomState(1),
  max_queue_len=8,
  early_stop_fn=no_progress_loss_custom(50,0) # Stops hyperopt tuning when loss does not improve after x iterations. Non custom version does not work with SparkTrials
)

How can I make fn=objective to accept the two additional positional arguments?

Solution

So long as x1 and x2 are specified at the fmin scope, you can use a lambda function:

fn=lambda space, x1=x1, x2=x2: objective(space, x1, x2),