BayesianOptimization 搜索不會因缺少參數而出錯

Question

以下代碼不會因 validation_data 中缺少參數而出錯

tuner_nn = BayesianOptimization(nn_builder,
                                objective = 'val_loss',
                                max_trials = 20,
                                seed = SEED,
                                directory = '/Users/Farid/Documents/kerastuner',
                                overwrite = True
                                )

tuner_nn.search(x_train, y_train, epochs=50, validation_data=(x_val, ), verbose=0, callbacks=[Earlystopping])

然而它錯誤validation_data=(x_val, y_val)

TypeError: 'float' object is not subscriptable

請參閱以下相關問題： typeerror-float-object-is-not-subscri

整行：

tuner_nn.search(x_train, y_train, epochs=50, validation_data=(x_val, y_val), verbose=0, callbacks=[Earlystopping])

Answer 1

我可以從您的鏈接中看到您的驗證數據的形狀是：

x_val.shape
>> (77, 1, 4)
y_val.shape
>> (77, 1)

也許嘗試將np.reshape()與您的 y_val 一起使用以將其轉換為 3D，這樣形狀將是 (77,1,1) 並查看是否可以解決問題：

y_val = y_val.reshape((y_val.shape[0], y_val.shape[1], 1))