I can't find any documentation for tf.app.flags, but I see that the command line parser will happily accept invalid syntax, flags that have never been defined, etc. Is there a way to configure it to raise an error in these cases? I wasted a lot of time trying to figure out why decreasing my learning rate didn't help when the problem was just that I had typed "-learning_rate" instead of "--learning_rate".
Asked
Active
Viewed 585 times
7
-
Doesn't really answer your question, but I use argparse from the python standard library instead, this being one of the reasons. – etarion Apr 14 '16 at 06:31
-
I wish TF used Google's gflags library rather than it's own manual parsing. This would also solve #1258 and and http://stackoverflow.com/questions/36331419/tensorflow-how-to-measure-how-much-gpu-memory-each-tensor-takes – Yaroslav Bulatov Apr 14 '16 at 17:01