AttributeError: 'tuple' object has no attribute 'fields' about pytorch-sentiment-analysis

[bentrevett]

The issue here is that TorchText doesn't like it when you only provide training data and no test/validation data. train_data is a one element tuple containing your TabularDataset, instead of just being the TabularDataset itself.

A quick fix is to change the following line:

train_data = torchtext.data.TabularDataset.splits(path = './', train = 'enron.csv', format = 'csv', fields = [('text',Text)])

We can just make it read the same dataset again as a test dataset, but then never actually use it:

train_data, _ = torchtext.data.TabularDataset.splits(path = './', train ='enron.csv', test ='enron.csv', format = 'csv', fields = [('text',Text)])

[Ravikiran2611]

THANK YOU SO MUCH @bentrevett
BUT NOW I GET AN ERROR FOR THE FOLLOWING LINE
for example in train_iterator: print(example)

AttributeError: 'Field' object has no attribute 'vocab'
IS IT BECAUSE I HAVE NOT EXECUTED THE LINE
Text.built_vocab(train_data)
IF SO , IS THERE ANY WAY TO SET THE VOCAB TO THE VOCAB I HAVE BECAUSE I AM USING A PRETRAINED BERT , THAT IS ,IS IT POSSIBLE TO NUMERICALIZE USING THE BERT TOKENIZER
PLS REPLY IF THERE ANY OTHER WAY TO CREATE DATASET AND CREATE BATCHES FOR IT

[bentrevett]

Yep, that's the reason for your error. Unfortunately TorchText doesn't have a nice way to load existing vocabularies (yet). See: pytorch/text#439 and pytorch/text#555