Bug?

[andreasvlachos]

I now got stuck here:

it reaches this point:

Training classifier<class 'type'>
Training classifier<class 'type'>
207
Embeddings: 208 x 300
Epoch 1/100

then:

2s - loss: 0.7413 - acc: 0.7379
Epoch 54/100
2s - loss: 0.7395 - acc: 0.7390
100%|████████████████████████████████████████████| 5/5 [12:27<00:00, 155.53s/it]
Epoch 55/100
2s - loss: 0.7394 - acc: 0.7385
Epoch 56/100

finally:

Epoch 99/100
2s - loss: 0.7136 - acc: 0.7441
Epoch 100/100
2s - loss: 0.7134 - acc: 0.7445
Traceback (most recent call last):
  File "fnc_ensemble.py", line 85, in <module>
    pickle.dump(slaves, open("features/slaves.pickle","wb+"))
_pickle.PicklingError: Can't pickle <class 'module'>: attribute lookup module on builtins failed

[j6mes]

yeah, something i'm working on. it may be best just evaluating without Mingjie's classifier for now (remove it from line 47 on fnc-ensemble.py)

[andreasvlachos]

Seems sorted.