我如何挑选我的神经网络预测模型，这样我就不必每次 re-train 它们了？

Question

正如标题所说，我想 pickle 我的 NN，但我得到了 TypeError

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_16896/2912615623.py in <module>
      5 
      6 with open('lstm_model.pkl', 'wb') as file:
----> 7     pickle.dump(mod, file)

TypeError: cannot pickle 'weakref' object

我尝试了 this SO question 的解决方案，但我遇到了同样的错误。
顺便说一句，该模型是来自 sklearn lib

的 LSTM

编辑：
抱歉，@Ben Reiniger 提到的 LSTM 来自 keras.layer

这是可重现的代码。

model = Sequential()
model.add(LSTM(50, return_sequences=True, input_shape = (x_train.shape[1], 1)))
model.add(LSTM(50, return_sequences = False))
model.add(Dense(25))
model.add(Dense(1))

model.compile(optimizer="adam", loss='mean_squared_error')

model.fit(x_train, y_train, batch_size = 1, epochs = 1)


# Saving the model
import pickle

mod = {'Model': model}

with open('lstm_model.pkl', 'wb') as file:
    pickle.dump(mod, file)

我得到的错误


---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_16896/2912615623.py in <module>
      5 
      6 with open('lstm_model.pkl', 'wb') as file:
----> 7     pickle.dump(mod, file)

TypeError: cannot pickle 'weakref' object

Answer 1

Keras 在 docs 中描述了一个 save 方法。他们说

it is not recommended to use pickle or cPickle to save a Keras model

因此，做：

节省：

model.save('your/path')

加载中：

model = keras.models.load_model('your/path')

我如何挑选我的神经网络预测模型，这样我就不必每次 re-train 它们了？

How do I pickle my neural net prediction models, so that i don't have to re-train them everytime?

python

neural-network

scikit-learn

deep-learning