train_classes =to_categorical(train_labels)
test_classes =to_categorical(test_labels)


score = {'AUC': 'roc_auc', 'Accuracy': make_scorer(accuracy_score)}
with tf.device('/device:GPU:0'):

    grid =  GridSearchCV(estimator = kerasmodel, param_grid = p, cv = 4, scoring= score,refit = 'AUC',verbose = 5)

    grid_result = (grid.fit(train_data, train_classes,shuffle = 'true'))

    print("Best: %f using %s" % (grid_result.best_score_,
    grid_result.best_params_))

    bestmodel = grid.best_estimator_

    preds = model.predict_proba(test_data)
    auc = roc_auc_score(test_classes, preds)
    acc = accuracy_score(test_classes[:,1], (preds[:,0]<preds[:,1])*1)

    preds2 = model.predict(test_data)
    auc2 = roc_auc_score(test_labels,preds2)

Dense(10, activation = activation),
    Dense(2, activation ='softmax')

model.compile(optimizer=optimizer(lr=lr),
                              loss = loss,
                              metrics = ['accuracy'])