I run the code of Meta-Learning with Memory-Augmented Neural Networks,but the result is not as predicted.
The accuracy of 1st instance is larger than 10th instance.Image results in the attachment.Why is this?
Is the order of 1st and 10th reversed?

when it is trying to train with the train_omniglot.py, i am hitting into this issue. can shed me some lights?

Traceback (most recent call last):
File "train_omniglot.py", line 52, in
for i, (images, labels) in train_generator:
File "../utils/generators.py", line 44, in next
return self.next()
File "../utils/generators.py", line 53, in next
for _img in zip(_images)]
File "../utils/generators.py", line 53, in
for _img in zip(_images)]
TypeError: The first input argument needs to be a sequence

I wonder which version of Chainer did you use？

This is a great project! Could you describe the intuition behind your data split procedure?
Obviously you mixed the original background and evaluation sets to create the new train and test sets.
I can also see that the created train and test sets don't match the setup in the paper 'One-shot Learning with Memory-Augmented Neural Networks'.

Thanks you!
T