Using tf.app.run() to call main function from a class - tensorflow

Question

The question is simple, I'm trying to use tf.app.run() to call the main function in a class. But, the following code gives me an error. Any help is appreciated.
import tensorflow as tf
import sys
# Where to find data
tf.app.flags.DEFINE_string('f1', '', 'feature 1')
tf.app.flags.DEFINE_string('f2', '', 'feature 2')
FLAGS = tf.app.flags.FLAGS
class Test(object):
def __init__(self):
pass
def main(self, args):
print(FLAGS.__flag.iteritems())
def main(args):
test = Test()
test.main(args)
if __name__ == '__main__':
tf.app.run(main)
And here is the error:
Traceback (most recent call last):
File "test.py", line 21, in <module>
tf.app.run(main)
File "/Users/yaserkeneshloo/anaconda/envs/env27/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "test.py", line 18, in main
test.main(args)
File "test.py", line 14, in main
print(FLAGS.__flag.iteritems())
File "/Users/yaserkeneshloo/anaconda/envs/env27/lib/python2.7/site-packages/tensorflow/python/platform/flags.py", line 85, in __getattr__
return wrapped.__getattr__(name)
File "/Users/yaserkeneshloo/anaconda/envs/env27/lib/python2.7/site-packages/absl/flags/_flagvalues.py", line 470, in __getattr__
raise AttributeError(name)
AttributeError: _Test__flag

Answer 1

The issue is that when you try to access a property of FLAGS beginning with double underscore from inside a class, it prefixes the property with the class name. So instead of FLAGS.__flags it tries to find FLAGS._Test__flags which does not exist.
This is true for both setting or getting a value. So if you set the value from outside of the class, you have to prefix it with _Test (because you named your class Test.) If you set the flags inside the class, you don't need to prefix, because it does the prefixing automatically also on assigning a value.
So basically you have no problem at all with your code because your own flags don't start with double underscore, except you cannot use the internal __flags property to print all flags. You can access them individually though.
See the code below for detailed examples. (Also, the default value in the DEFINE line is the second parameter, not the third.)
# Where to find data
tf.app.flags.DEFINE_string('_Test__f1', 'feature 1', 'feature 1')
tf.app.flags.DEFINE_string('__f2', 'feature 2', 'feature 2')
FLAGS = tf.app.flags.FLAGS
print( FLAGS.__f2 ) # prints "feature 2"
class Test(object):
def __init__(self):
pass
def main(self, args):
print( FLAGS.__f1 ) # prints "feature 1"
FLAGS.__f1 = 'foobar' # assignment works the same way
print( FLAGS.__f1 ) # prints "foobar"
print( FLAGS.__f2 ) # AttributeError: _Test__f2
def main(args):
test = Test()
test.main(args)
if __name__ == '__main__':
tf.app.run(main)

Answer 2

The problem is that FLAGS doesn't have an attribute named __flag. If you want to print the string corresponding to the f1 flag, call print(FLAGS.f1).

Answer 3

This will do the trick:
import tensorflow as tf
import sys
# Where to find data
tf.app.flags.DEFINE_string('f1', '', 'feature 1')
tf.app.flags.DEFINE_string('f2', '', 'feature 2')
FLAGS = tf.app.flags.FLAGS
class Test(object):
def __init__(self):
pass
def main(self, args):
flags = getattr(FLAGS,"__flags")
print([(k,v) for (k,v) in flags.iteritems()])
def main(args):
test = Test()
test.main(args)
if __name__ == '__main__':
tf.app.run(main)