Keras fit_generator Validation Data TypeError: 'float' object cannot be interpreted as an integer

Question

I'm trying to run CUHK03 Person Re-ID script by Ning-Ding ( Implementation of Ahmed et. al.'s paper using Keras) see https://github.com/Ning-Ding/Implementation-CVPR2015-CNN-for-ReID

Error text reads as follows:

TypeError Traceback (most recent call last)
in ()

    ----> 1 main("E:\DL\cuhk-03.h5")

in main(dataset_path)

    17 model = generate_model()
    18 model = compile_model(model)
    ---> 19 train(model, dataset_path)
    20
    21 def train(model,

in train(model, h5_path, weights_name, train_num, one_epoch, epoch_num, flag_random, random_pattern, flag_train, flag_val, which_val_data, nb_val_samples)
    39 rand_x = np.random.rand()
    40 flag_train = random_pattern(rand_x)
    ---> 41 model.fit_generator(Data_Generator.flow(f,flag = flag_train),one_epoch,epoch_num,validation_data=Data_Generator.flow(f,train_or_validation=which_val_data,flag=flag_val),nb_val_samples=nb_val_samples)
    42 Rank1s.append(round(cmc(model)[0],2))
    43 print (Rank1s)

~\Anaconda3\lib\site-packages\keras\legacy\interfaces.py in wrapper(*args, **kwargs)

    89 warnings.warn('Update your ' + object_name + 90 ' call to the Keras 2 API: ' + signature, stacklevel=2)
    ---> 91 return func(*args, **kwargs)
    92 wrapper._original_function = func
    93 return wrapper

~\Anaconda3\lib\site-packages\keras\engine\training.py in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)

    2023 epoch = initial_epoch
    2024
    -> 2025 do_validation = bool(validation_data)
    2026 self._make_train_function()
    2027 if do_validation:

TypeError: 'float' object cannot be interpreted as an integer

I am using Jupyter Notebook in Anaconda on Windows 10(x86). Keras version 2.1.3 Python version 3.6.3 Tensorflow backend (1.4.0)

Answer 1

Ok, so that validation_data is a generator returned by

Data_Generator.flow(f,train_or_validation=which_val_data,flag=flag_val)

When do_validation = bool(validation_data) is executed, calling bool on an object will call nonzero or len if any of them is defined. In this case, Sequence implements len so it checks if len(Sequence) == 0. Your issue is that len returns a float (which is an error) so when it tries to convert it in a bool, it fails.

Assert that len returns an int.

Credit goes to Dref360 at https://www.bountysource.com/issues/54744813-fit_generator-throws-error-on-validation-data-being-float-data-type