CODEWITHSUNDEEP
pythongeneratorkeras
A_Matar
A_Matar

Reputation: 2320

Model.fit: when you are using validation_generator you must specify a value for validation_steps

I am getting this error when I try to compile a python script that uses Keras

ValueError                                Traceback (most recent call last)
/home/cse/abdelrahmanML/project/cervix/cervixXception.py in <module>()
    160                         validation_steps=len(valid_list)//conf['batch_size'],
    161                         verbose=1,
--> 162             callbacks=myCallbacks)
    163 
    164 

/home/cse/venv/local/lib/python2.7/site-packages/keras/legacy/interfaces.pyc in wrapper(*args, **kwargs)
     85                 warnings.warn('Update your `' + object_name +
     86                               '` call to the Keras 2 API: ' + signature)
---> 87             return func(*args, **kwargs)
     88         return wrapper
     89     return legacy_support

/home/cse/venv/local/lib/python2.7/site-packages/keras/engine/training.pyc in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_q_size, workers, pickle_safe, initial_epoch)
   1781                    hasattr(validation_data, '__next__'))
   1782         if val_gen and not validation_steps:
-> 1783             raise ValueError('When using a generator for validation data, '
   1784                              'you must specify a value for '
   1785                              '`validation_steps`.')

ValueError: When using a generator for validation data, you must specify a value for `validation_steps`.

Here is my piece of code that generates the error, as you can notice it does specify a value for validation_steps. I can't find out what is wrong:

fit = model.fit_generator(generator=batch_generator_train(train_list, conf['batch_size']),
            steps_per_epoch=len(train_list)//conf['batch_size'],
            nb_epoch=conf['nb_epoch'],
            validation_data=batch_generator_train(valid_list, conf['batch_size']),
            validation_steps=len(valid_list)//conf['batch_size'],
            verbose=1,
            callbacks=myCallbacks)

Note that: 

conf = dict()
conf['batch_size'] = 16

Upvotes: 2

Views: 2572

Answers (1)

Colwin
Colwin

Reputation: 2685

Okay second attempt at answering this.

After looking at the source for training.py I can only see one scenario when this will occur when len(valid_list) is less than 16 (the value of batch_size). This will cause the floor division to return 0 and cause that if to pass and raise that error you are seeing.

Upvotes: 4

Related Questions