Optuna LSTM

lstm-optuna.py

import optuna
from optuna.trial import TrialState
from sklearn.metrics import accuracy_score

def objective(trial):
  optimizer_name = trial.suggest_categorical("optimizer", ["adam", "SGD", "RMSprop", "Adadelta"])
  epochs = trial.suggest_int("epochs", 5, 15,step=5, log=False)
  batchsize = trial.suggest_int("batchsize", 8, 40,step=16, log=False)

  history, model = lstm(optimizer_name,epochs,batchsize)

  val_acc = model.evaluate(X_val,y_val)[1]
  weights = model.get_weights()

  # Handle pruning based on the intermediate value.
  if trial.should_prune():
    raise optuna.exceptions.TrialPruned()

  trial.set_user_attr(key="best_model_weights", value=weights)
  return val_acc

def callback(study, trial):
    if study.best_trial.number == trial.number:
        study.set_user_attr(key="best_model_weights", 
                            value=trial.user_attrs["best_model_weights"])

study = optuna.create_study(direction="maximize")
study.optimize(objective, n_trials=20, timeout=None, callbacks=[callback])