代码实现（4） Matching Networks

1. 运行

1.1 运行环境

(1) Pytorch
(2) tqdm
(3) tensorboard_logger

1.2 问题

Traceback (most recent call last):
  File "C:/Users/xpb/PycharmProjects/MatchingNetworks-pytorch/mainOmniglot.py", line 47, in <module>
    samples_per_class=samples_per_class)
  File "C:\Users\xpb\PycharmProjects\MatchingNetworks-pytorch\datasets\omniglotNShot.py", line 45, in __init__
    self.x.append(np.array(temp[temp.keys()[classes]]))
TypeError: 'dict_keys' object does not support indexing

问题原因： 这是由于python3改变了dict.keys,返回的是dict_keys对象,支持iterable 但不支持indexable，我们可以将其明确的转化成list：
解决方法：  
self.x.append(np.array(temp[list(temp.keys())[classes]]))

2.3 memory error

内存太小
for sample in range(100):

2.4

Traceback (most recent call last):
  File "C:/Users/xpb/PycharmProjects/MatchingNetworks-pytorch/mainOmniglot.py", line 55, in <module>
    total_c_loss, total_accuracy = obj_oneShotBuilder.run_training_epoch(total_train_batches=total_train_batches)
  File "C:\Users\xpb\PycharmProjects\MatchingNetworks-pytorch\experiments\OneShotBuilder.py", line 115, in run_training_epoch
    iter_out = "tr_loss: {}, tr_accuracy: {}".format(c_loss_value.data[0], acc.data[0])
IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number

解决方法：
  #  iter_out = "tr_loss: {}, tr_accuracy: {}".format(c_loss_value.data[0], acc.data[0])
   iter_out = "tr_loss: {}, tr_accuracy: {}".format(c_loss_value.item(), acc.item())
        
# total_c_loss += c_loss_value.data[0]
 # total_accuracy += acc.data[0]
 total_c_loss += c_loss_value.item()
total_accuracy += acc.item()       

参考资料

[1] tensorboard_logger

问题解决

[1] TypeError: 'dict_keys' object does not support indexing

代码

[1] AntreasAntoniou/MatchingNetworks
[2] gitabcworld/MatchingNetworks
[3] cnichkawde/MatchingNetwork

论文

[1] Matching Networks for One Shot Learning