model = model_util.load(model_to_load, filepath)
#To prevent any possible corruption between cycles, just reload and
#reprocess the data, including resorting
data, classes = emnist.get(balanced_set)
x, y, val_x, val_y = data_util.validation_split(data['x'], data['y'], validation_split)
x, y = emnist.mean_sort(x, y, classes, task_count)
#Convert test data to be used in model
val_x, val_y = data_util.prep(val_x, val_y, classes)
test_x, test_y = data_util.prep(data['test_x'], data['test_y'], classes)
#Set up syllabus which handles splitting into tasks
syllabus = Syllabus(
training_data =(x,y),
epochs=epochs,
task_amount=task_count,
weightings=distribution,
)
#Until we have finished training
while not syllabus.training_complete():
#Get our next task and prep it
x, y = syllabus.next()
x, y = data_util.prep(x, y, classes)
print("epoch : {0}\ttask: {1}".format(syllabus.current_epoch, syllabus.current_task_index))
model.fit(
x=x,
y=y,
batch_size=batch_size,
epochs=1,
