使用pickle保存hyperopt Trials字典时仅trial_0留存，trial_1丢失如何解决？

我在使用pickle保存hyperopt的Trials对象字典时遇到问题：
我的trials字典包含两个Trials对象：

trials = {
    'trial_0': <hyperopt.base.Trials object at 0x0000020B6F875E50>,
    'trial_1': <hyperopt.base.Trials object at 0x0000020C32DA9490>
}

执行以下代码进行保存：

fname = r'C:\Users\test09.pkl'
with open(fname, 'wb+') as fpkl:
    pass
for trial_label, trial in trials.items():
    print(f"\nData for {trial_label}:")
    with open(fname, "ab") as file:
        pickle.dump(trials[trial_label], file)

通过print可确认两个Trials对象都有完整数据，但读取pkl文件时（代码如下）仅能获取trial_0的内容，trial_1的信息丢失：

with open(r'C:\Users\test09.pkl','rb') as file:
    data = pickle.load(file)

for trial in data.trials:
    print(trial)

with open(r'C:\Users\output_test10.txt', 'w') as file:
     for alle in data.trials:
        file.write(str(alle)+'\n')

请问为何trial_1的信息未被写入pkl文件？该如何解决此问题？

不是trial_1没写入文件，而是读取方式有误。你分两次调用pickle.dump()，会把两个Trials对象依次序列化到同一个pkl文件中，但读取时只调用了一次pickle.load()，该方法仅会读取文件里的第一个序列化对象（即trial_0），第二个对象仍留在文件中未被读取，所以你看不到trial_1的内容。

方法一：直接保存整个字典（推荐） #

没必要分开保存每个Trials对象，直接把整个trials字典一次性dump到文件，读取时直接加载整个字典即可获取两个对象：

# 保存
fname = r'C:\Users\test09.pkl'
with open(fname, 'wb') as file:
    pickle.dump(trials, file)

# 读取
with open(fname, 'rb') as file:
    loaded_trials = pickle.load(file)

# 访问两个对象
print(loaded_trials['trial_0'].trials)
print(loaded_trials['trial_1'].trials)

方法二：读取所有序列化对象 #

如果必须分开dump，读取时需要循环调用pickle.load()直到文件指针到达末尾：

# 读取
loaded_trials = []
with open(r'C:\Users\test09.pkl','rb') as file:
    while True:
        try:
            trial = pickle.load(file)
            loaded_trials.append(trial)
        except EOFError:
            break  # 读到文件末尾，停止循环

# loaded_trials[0]是trial_0，loaded_trials[1]是trial_1
print(loaded_trials[0].trials)
print(loaded_trials[1].trials)

内容的提问来源于stack exchange，提问作者Lance Zhang