关于Ryan Dahl的tensorflow-resnet中Co

本文旨在学习Ryan Dahl的tensorflow-resnet源码中的Config类的基本作用。因为，它真的真的很有趣！

tensorflow-resnet的repository中有个文件叫config.py，Config类就是在这个文件中被定义的。它能够很方便的实现基于tensorflow编写程序的不同参数在不同scope中的隔离管理（很拗口，一会儿上例子），Config类有以下几个特点：

• 它可以被认为是包含了多个dict的list
• 它的内部参数在不同variable scope中是“隔离”的

说了那么多，相信谁都没看明白，那么举个栗子：

c = Config()
c['p1'] = 1
c['p2'] = 1
c['p3'] = 1
# c['p1'] = 1, c['p2'] = 1, c['p3'] = 1, c['p4']不存在

with tf.variable_scope('foo'):
    c['p1'] = 2
    c['p4'] = 2
    # c['p1'] = 2, c['p2'] = 1, c['p3'] = 1, c['p4'] = 2

    with tf.variable_scope('bar'):
        c['p2'] = 2
        # c['p1'] = 2, c['p2'] = 2, c['p3'] = 1, c['p4'] = 2

with tf.variable_scope('baz'):
    c['p3'] = 2
    # c['p1'] = 1, c['p2'] = 1, c['p3'] = 2, c['p4']不存在

# c['p1'] = 1, c['p2'] = 1, c['p3'] = 1, c['p4']不存在

程序内各项参数在不同位置的取值我已经注释出来了，很明显，不同variable scope中的参数是隔离的，你在’foo‘中设置的参数在’baz‘不起作用，在’foo‘中新定义的参数在其他scope中看不到（但在’foo‘中的’bar‘内可以看到）。

关于Config类的特点还有待挖掘，以上只是说明了它最基本的特点，下面给出Ryan Dahl的config.py的源码，你也可以去他的repository中看，这是链接。

# This is a variable scope aware configuation object for TensorFlow

import tensorflow as tf

FLAGS = tf.app.flags.FLAGS

class Config:
    def __init__(self):
        root = self.Scope('')
        for k, v in FLAGS.__dict__['__flags'].iteritems():
            root[k] = v
        self.stack = [ root ]

    def iteritems(self):
        return self.to_dict().iteritems()

    def to_dict(self):
        self._pop_stale()
        out = {}
        # Work backwards from the flags to top fo the stack
        # overwriting keys that were found earlier.
        for i in range(len(self.stack)):
            cs = self.stack[-i]
            for name in cs:
                out[name] = cs[name]
        return out

    def _pop_stale(self):
        var_scope_name = tf.get_variable_scope().name
        top = self.stack[0]
        while not top.contains(var_scope_name):
            # We aren't in this scope anymore
            self.stack.pop(0)
            top = self.stack[0]

    def __getitem__(self, name):
        self._pop_stale()
        # Recursively extract value
        for i in range(len(self.stack)):
            cs = self.stack[i]
            if name in cs:
                return cs[name]

        raise KeyError(name)

    def set_default(self, name, value):
        if not (name in self):
            self[name] = value

    def __contains__(self, name):
        self._pop_stale()
        for i in range(len(self.stack)):
            cs = self.stack[i]
            if name in cs:
                return True
        return False

    def __setitem__(self, name, value):
        self._pop_stale()
        top = self.stack[0]
        var_scope_name = tf.get_variable_scope().name
        assert top.contains(var_scope_name)

        if top.name != var_scope_name:
            top = self.Scope(var_scope_name)
            self.stack.insert(0, top)

        top[name] = value

    class Scope(dict):
        def __init__(self, name):
            self.name = name

        def contains(self, var_scope_name):
            return var_scope_name.startswith(self.name)



# Test
if __name__ == '__main__':

    def assert_raises(exception, fn):
        try:
            fn()
        except exception:
            pass
        else:
            assert False, "Expected exception"

    c = Config()

    c['hello'] = 1
    assert c['hello'] == 1

    with tf.variable_scope('foo'):
        c.set_default("bar", 10)
        c['bar'] = 2
        assert c['bar'] == 2
        assert c['hello'] == 1

        c.set_default("mario", True)

        with tf.variable_scope('meow'):
            c['dog'] = 3
            assert c['dog'] == 3
            assert c['bar'] == 2
            assert c['hello'] == 1

            assert c['mario'] == True

        assert_raises(KeyError, lambda: c['dog'])
        assert c['bar'] == 2
        assert c['hello'] == 1