Python中的数据对象持久化存储模块pickle的使用示例

1054次阅读  |  发布于5年以前

Python中可以使用 pickle 模块将对象转化为文件保存在磁盘上,在需要的时候再读取并还原。具体用法如下:
pickle是Python库中常用的序列化工具,可以将内存对象以文本或二进制格式导出为字符串,或者写入文档。后续可以从字符或文档中还原为内存对象。新版本的Python中用c重新实现了一遍,叫cPickle,性能更高。 下面的代码演示了pickle库的常用接口用法,非常简单:


    import cPickle as pickle

    # dumps and loads
    # 将内存对象dump为字符串,或者将字符串load为内存对象
    def test_dumps_and_loads():
      t = {'name': ['v1', 'v2']}
      print t

      o = pickle.dumps(t)
      print o
      print 'len o: ', len(o)

      p = pickle.loads(o)
      print p



    # 关于HIGHEST_PROTOCOL参数,pickle 支持3种protocol,0、1、2:
    # http://stackoverflow.com/questions/23582489/python-pickle-protocol-choice
    # 0:ASCII protocol,兼容旧版本的Python
    # 1:binary format,兼容旧版本的Python
    # 2:binary format,Python2.3 之后才有,更好的支持new-sytle class
    def test_dumps_and_loads_HIGHEST_PROTOCOL():
      print 'HIGHEST_PROTOCOL: ', pickle.HIGHEST_PROTOCOL

      t = {'name': ['v1', 'v2']}
      print t

      o = pickle.dumps(t, pickle.HIGHEST_PROTOCOL)
      print 'len o: ', len(o)

      p = pickle.loads(o)
      print p


    # new-style class
    def test_new_sytle_class():
      class TT(object):
        def __init__(self, arg, **kwargs):
          super(TT, self).__init__()
          self.arg = arg
          self.kwargs = kwargs

        def test(self):
          print self.arg
          print self.kwargs

      # ASCII protocol
      t = TT('test', a=1, b=2)
      o1 = pickle.dumps(t)
      print o1
      print 'o1 len: ', len(o1)
      p = pickle.loads(o1)
      p.test()

      # HIGHEST_PROTOCOL对new-style class支持更好,性能更高
      o2 = pickle.dumps(t, pickle.HIGHEST_PROTOCOL)
      print 'o2 len: ', len(o2)
      p = pickle.loads(o2)
      p.test()


    # dump and load
    # 将内存对象序列化后直接dump到文件或支持文件接口的对象中
    # 对于dump,需要支持write接口,接受一个字符串作为输入参数,比如:StringIO
    # 对于load,需要支持read接口,接受int输入参数,同时支持readline接口,无输入参数,比如StringIO

    # 使用文件,ASCII编码
    def test_dump_and_load_with_file():
      t = {'name': ['v1', 'v2']}

      # ASCII format
      with open('test.txt', 'w') as fp:
        pickle.dump(t, fp)

      with open('test.txt', 'r') as fp:
        p = pickle.load(fp)
        print p


    # 使用文件,二进制编码
    def test_dump_and_load_with_file_HIGHEST_PROTOCOL():
      t = {'name': ['v1', 'v2']}
      with open('test.bin', 'wb') as fp:
        pickle.dump(t, fp, pickle.HIGHEST_PROTOCOL)

      with open('test.bin', 'rb') as fp:
        p = pickle.load(fp)
        print p


    # 使用StringIO,二进制编码
    def test_dump_and_load_with_StringIO():
      import StringIO

      t = {'name': ['v1', 'v2']}

      fp = StringIO.StringIO()
      pickle.dump(t, fp, pickle.HIGHEST_PROTOCOL)

      fp.seek(0)
      p = pickle.load(fp)
      print p

      fp.close()


    # 使用自定义类
    # 这里演示用户自定义类,只要实现了write、read、readline接口,
    # 就可以用作dump、load的file参数
    def test_dump_and_load_with_user_def_class():
      import StringIO

      class FF(object):
        def __init__(self):
          self.buf = StringIO.StringIO()

        def write(self, s):
          self.buf.write(s)
          print 'len: ', len(s)

        def read(self, n):
          return self.buf.read(n)

        def readline(self):
          return self.buf.readline()

        def seek(self, pos, mod=0):
          return self.buf.seek(pos, mod)

        def close(self):
          self.buf.close()

      fp = FF()
      t = {'name': ['v1', 'v2']}
      pickle.dump(t, fp, pickle.HIGHEST_PROTOCOL)

      fp.seek(0)
      p = pickle.load(fp)
      print p

      fp.close()


    # Pickler/Unpickler
    # Pickler(file, protocol).dump(obj) 等价于 pickle.dump(obj, file[, protocol])
    # Unpickler(file).load() 等价于 pickle.load(file)
    # Pickler/Unpickler 封装性更好,可以很方便的替换file
    def test_pickler_unpickler():
      t = {'name': ['v1', 'v2']}

      f = file('test.bin', 'wb')
      pick = pickle.Pickler(f, pickle.HIGHEST_PROTOCOL)
      pick.dump(t)
      f.close()

      f = file('test.bin', 'rb')
      unpick = pickle.Unpickler(f)
      p = unpick.load()
      print p
      f.close()

pickle.dump(obj, file[, protocol])
这是将对象持久化的方法,参数的含义分别为:

对象被持久化后怎么还原呢?pickle 模块也提供了相应的方法,如下:

pickle.load(file)
只有一个参数 file ,对应于上面 dump 方法中的 file 参数。这个 file 必须是一个拥有一个能接收一个整数为参数的 read() 方法以及一个不接收任何参数的 readline() 方法,并且这两个方法的返回值都应该是字符串。这可以是一个打开为读的文件对象、StringIO 对象或其他任何满足条件的对象。

下面是一个基本的用例:


    # -*- coding: utf-8 -*-

    import pickle
    # 也可以这样:
    # import cPickle as pickle

    obj = {"a": 1, "b": 2, "c": 3}

    # 将 obj 持久化保存到文件 tmp.txt 中
    pickle.dump(obj, open("tmp.txt", "w"))

    # do something else ...

    # 从 tmp.txt 中读取并恢复 obj 对象
    obj2 = pickle.load(open("tmp.txt", "r"))

    print obj2

    # -*- coding: utf-8 -*-

    import pickle
    # 也可以这样:
    # import cPickle as pickle

    obj = {"a": 1, "b": 2, "c": 3}

    # 将 obj 持久化保存到文件 tmp.txt 中
    pickle.dump(obj, open("tmp.txt", "w"))

    # do something else ...

    # 从 tmp.txt 中读取并恢复 obj 对象
    obj2 = pickle.load(open("tmp.txt", "r"))

    print obj2

不过实际应用中,我们可能还会有一些改进,比如用 cPickle 来代替 pickle ,前者是后者的一个 C 语言实现版本,拥有更快的速度,另外,有时在 dump 时也会将第三个参数设为 True 以提高压缩比。再来看下面的例子:


    # -*- coding: utf-8 -*-

    import cPickle as pickle
    import random
    import os

    import time

    LENGTH = 1024 * 10240

    def main():
     d = {}
     a = []
     for i in range(LENGTH):
     a.append(random.randint(0, 255))

     d["a"] = a

     print "dumping..."

     t1 = time.time()
     pickle.dump(d, open("tmp1.dat", "wb"), True)
     print "dump1: %.3fs" % (time.time() - t1)

     t1 = time.time()
     pickle.dump(d, open("tmp2.dat", "w"))
     print "dump2: %.3fs" % (time.time() - t1)

     s1 = os.stat("tmp1.dat").st_size
     s2 = os.stat("tmp2.dat").st_size

     print "%d, %d, %.2f%%" % (s1, s2, 100.0 * s1 / s2)

     print "loading..."

     t1 = time.time()
     obj1 = pickle.load(open("tmp1.dat", "rb"))
     print "load1: %.3fs" % (time.time() - t1)

     t1 = time.time()
     obj2 = pickle.load(open("tmp2.dat", "r"))
     print "load2: %.3fs" % (time.time() - t1)


    if __name__ == "__main__":
     main()

    # -*- coding: utf-8 -*-

    import cPickle as pickle
    import random
    import os

    import time

    LENGTH = 1024 * 10240

    def main():
     d = {}
     a = []
     for i in range(LENGTH):
     a.append(random.randint(0, 255))

     d["a"] = a

     print "dumping..."

     t1 = time.time()
     pickle.dump(d, open("tmp1.dat", "wb"), True)
     print "dump1: %.3fs" % (time.time() - t1)

     t1 = time.time()
     pickle.dump(d, open("tmp2.dat", "w"))
     print "dump2: %.3fs" % (time.time() - t1)

     s1 = os.stat("tmp1.dat").st_size
     s2 = os.stat("tmp2.dat").st_size

     print "%d, %d, %.2f%%" % (s1, s2, 100.0 * s1 / s2)

     print "loading..."

     t1 = time.time()
     obj1 = pickle.load(open("tmp1.dat", "rb"))
     print "load1: %.3fs" % (time.time() - t1)

     t1 = time.time()
     obj2 = pickle.load(open("tmp2.dat", "r"))
     print "load2: %.3fs" % (time.time() - t1)


    if __name__ == "__main__":
     main()

在我的电脑上执行结果为:


    dumping…
    dump1: 1.297s
    dump2: 4.750s
    20992503, 68894198, 30.47%
    loading…
    load1: 2.797s
    load2: 10.125s

可以看到,dump 时如果指定了 protocol 为 True,压缩过后的文件的大小只有原来的文件的 30% ,同时无论在 dump 时还是 load 时所耗费的时间都比原来少。因此,一般来说,可以建议把这个值设为 True 。

另外,pickle 模块还提供 dumps 和 loads 两个方法,用法与上面的 dump 和 load 方法类似,只是不需要输入 file 参数,输入及输出都是字符串对象,有些场景中使用这两个方法可能更为方便。

Copyright© 2013-2020

All Rights Reserved 京ICP备2023019179号-8