Keras: Getting NotImplementedError for "on_epoch_end" while training using fit_generator on a Sequence object

[rohit-gupta]

Exception in thread Thread-8:
Traceback (most recent call last):
File "/users/gpu/rohitg1/miniconda2/envs/tensorflow/lib/python2.7/threading.py", line 801, in __bootstrap_inner
self.run()
File "/users/gpu/rohitg1/miniconda2/envs/tensorflow/lib/python2.7/threading.py", line 754, in run
self.__target(self.__args, *self.__kwargs)
File "/users/gpu/rohitg1/.local/lib/python2.7/site-packages/keras/utils/data_utils.py", line 492, in _run
self.sequence.on_epoch_end()
File "/users/gpu/rohitg1/.local/lib/python2.7/site-packages/keras/utils/data_utils.py", line 358, in on_epoch_end
raise NotImplementedError
NotImplementedError

I am trying to train a Video classification model. Gist for portion of code is available here:

https://gist.github.com/rohit-gupta/7668b79389e29598ace41813fc1a50d2

Do I need to implement a "on_epoch_end" function in my Sequence object to get this o work ?

[Dref360]

yes you need to,
You can do it as simple as :

def on_epoch_end(self):
        pass

@joeyearsley Should we make it optional or clarify the doc instead to avoid issues for newcomers?

[joeyearsley]

I'ld say optional as it isn't always needed

[rohit-gupta]

@Dref360 Thanks for your reply, Its solved my issue ! Though I am curious as to what the purpose of this is ? What do people usually implement in this function ?

[Dref360]

If you want to shuffle your data between epoch, on_epoch_end is called at this time.

[fredtcaroli]

Had to read keras' code to figure out why my script was raising this NotImplementedError. This should definitely be documented or made optional asap

[jlherren]

Seems fixed with PR #8007

[fchollet]

Yes, it is fixed at master.