在Python中无法pickle Tensorflow对象 - TypeError: 无法pickle _thread._local对象。

Question

我想在使用TensorFlow的Keras fit后pickle历史对象，但是出现了错误。

import gzip
import numpy as np
import os
import pickle
import tensorflow as tf
from tensorflow import keras


with gzip.open('mnist.pkl.gz', 'rb') as f:
    train_set, test_set = pickle.load(f, encoding='latin1')

X_train = np.asarray(train_set[0])
y_train = np.asarray(train_set[1])

X_test = np.asarray(test_set[0])
y_test = np.asarray(test_set[1])

X_valid, X_train = X_train[:5000]/255.0, X_train[5000:]/255.0
y_valid, y_train = y_train[:5000], y_train[5000:]

class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle Boot']

model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28,28]))
model.add(keras.layers.Dense(300, activation = 'relu'))
model.add(keras.layers.Dense(100, activation = 'relu'))
model.add(keras.layers.Dense(10, activation = 'softmax'))
model.summary()

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

history = model.fit(X_train, y_train, epochs=1,
                    validation_data =(X_valid, y_valid))

if not os.path.isdir('models'):
    os.mkdir('models')

model.save('models/basic.h5')
with open('models/basic_history.pickle', 'wb') as f:
    pickle.dump(history, f)

它给我以下错误：

Traceback (most recent call last):
  File "main.py", line 69, in <module>
    pickle.dump(history, f)
TypeError: can't pickle _thread._local objects

注意：要运行代码，请下载时尚MNIST数据：https://s3.amazonaws.com/img-datasets/mnist.pkl.g

- Al-Baraa El-Hag

---

这个回答解决了你的问题吗？TypeError：无法pickle _thread.lock对象 - Reznik

或许尝试使用 history.history 来代替... - Karl

你试过使用dill了吗？或者那也出现了问题？ - Charlie Parker

Answer 1

正如Karl所建议的，history对象无法被序列化。但是它的字典可以：

with open('models/basic_history.pickle', 'wb') as f:
    pickle.dump(history.history, f)

- Al-Baraa El-Hag

Answer 2

Joblib 对我也起到了作用:

import joblib
model_filename = "lstm.pkl"
joblib.dump(history.history, model_filename)

- Pavel Fedotov