Tensorflow 打印所有tf.flags 参数的方法

迷南。 2022-03-12 02:24 313阅读 0赞

深度学习离不开大量的参数配置，Tensorflow内置的参数方法 tf.flags 是很多TF使用者的选择

我在设置参数之后想要每次运行程序时打印出所有参数用来核对与记录，费了些功夫找到了方法

不多说，上代码

import tensorflow as tf
    
    Flags=tf.flags
    
    Flags.DEFINE_float('learning_rate', 0.0001, 'The learning rate for the network')
    Flags.DEFINE_integer('decay_step', 500000, 'The steps needed to decay the learning rate')
    Flags.DEFINE_float('decay_rate', 0.1, 'The decay rate of each decay step')
    Flags.DEFINE_string('mode','train', 'The mode of the model train, test.')
    
    FLAGS = Flags.FLAGS
    
    def print_configuration_op(FLAGS):
        print('My Configurations:')
        #pdb.set_trace()
        for name, value in FLAGS.__flags.items():
            value=value.value
            if type(value) == float:
                print(' %s:\t %f'%(name, value))
            elif type(value) == int:
                print(' %s:\t %d'%(name, value))
            elif type(value) == str:
                print(' %s:\t %s'%(name, value))
            elif type(value) == bool:
                print(' %s:\t %s'%(name, value))
            else:
                print('%s:\t %s' % (name, value))
        #for k, v in sorted(FLAGS.__dict__.items()):
            #print(f'{k}={v}\n')        
        print('End of configuration')
    
    def main(argv):
        print_configuration_op(FLAGS)
    
    
    if __name__ == '__main__':
        tf.app.run()

输出

My Configurations:
     learning_rate:	 0.000100
     decay_step:	 500000
     decay_rate:	 0.100000
     mode:	 train
     h:	 False
     help:	 False
     helpfull:	 False
     helpshort:	 False
    End of configuration