Tried to construct a tensor from a int sequence, but found an item of type float at index (0)

hafidz · February 5, 2018, 7:28pm

Was trying to fit an LSTM model that is using a pretrained language model for the Kaggle Toxic competition. However currently I’m hitting this error, and I have no clue how to proceed next. Was wondering if anyone has hit this kind of error before

gist.github.com

https://gist.github.com/ikanez/ea4224f46102270aaf40e77396179a72

found an item of type float at index (0)

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-37-16682764c468> in <module>()
----> 1 fit(m, md, 1, lo.opt, F.binary_cross_entropy)
      2 # use F.binary_cross_entropy for multi-label problems

~\Dropbox\3.SelfStudy\fastai_pytorch\fastai\courses\dl1\fastai\model.py in fit(model, data, epochs, opt, crit, metrics, callbacks, **kwargs)
    104             i += 1
    105 
--> 106         vals = validate(stepper, data.val_dl, metrics)

This file has been truncated. show original

hafidz · March 3, 2018, 6:24pm

Update on the issue. A probable solution as mentioned by a commenter here (RuntimeError: tried to construct a tensor from a int sequence, but found an item of type numpy.int64 at index (0) · Issue #2 · facebookresearch/InferSent · GitHub)

When you define a Field in torchtext by default it is torch.LongTensor , so you can pass whatever type you want like this data.Field(sequential=False, use_vocab=False,tensor_type=torch.cuda.FloatTensor)

So happens when I changed my code accordingly as per his suggestion (below), the error no longer appears.
tt_TEXT = data.Field(sequential=True, tokenize=tokenizer, fix_length=max_len)
tt_LABEL = data.Field(sequential=False, use_vocab=False,tensor_type=torch.cuda.FloatTensor)