predict analysis

[roiDaniela]

I added this code to try predict a sentence:

`def predict_sentiment(model, sentence, dataset):

  nlp = spacy.load('en_core_web_sm')
  dataset = MyDataset(batch_size=1)
  tokenized = [tok.text for tok in nlp.tokenizer(sentence)]
  indexed = [dataset.TEXT.vocab.stoi[t] for t in tokenized]
  tensor = torch.LongTensor(indexed).to(device)
  tensor = tensor.unsqueeze(1)
  prediction = F.sigmoid(model(tensor))
  return prediction.item()`

`if name == 'main':

args = parse_args()
device = torch.device('cuda' if torch.cuda.is_available() and args.cuda else 'cpu')
dataset = MyDataset(batch_size=args.batch_size, use_vector=args.word_vector, pdevice = device)
print(args)
model = main(args, dataset, device)

print(predict_sentiment(model, " smart , provocative and blisteringly funny", dataset))`

but I got the error File "C:\Users\roifo\PycharmProjects\pytorch-sentiment-analysis-classification\model.py", line 101, in attention merged_state = merged_state.squeeze(0).unsqueeze(2) IndexError: Dimension out of range (expected to be in range of [-2, 1], but got 2)

I tried to change the line
tensor = tensor.unsqueeze(1) to tensor = tensor.unsqueeze(0) but then the error was warnings.warn("nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.") Traceback (most recent call last): File "C:/Users/roifo/PycharmProjects/pytorch-sentiment-analysis-classification/train.py", line 237, in <module> print(predict_sentiment(model, " smart , provocative and blisteringly funny", dataset)) File "C:/Users/roifo/PycharmProjects/pytorch-sentiment-analysis-classification/train.py", line 226, in predict_sentiment return prediction.item() ValueError: only one element tensors can be converted to Python scalars

thanks

[roiDaniela]

version details:

In [2]: torchtext.version Out[2]: '0.10.0'

In [2]: torch.version Out[2]: '1.9.0'

@slaysd thanks!

[roiDaniela]

did the change you set in requirements, still got error when running lstm_attn:

merged_state = merged_state.squeeze(0).unsqueeze(2) IndexError: Dimension out of range (expected to be in range of [-2, 1], but got 2)

[DenChenn]

@roiDaniela @slaysd

I have solved this function by the following code :

def predict_class(text, model, sentence):
    model.eval()
    tokenized = [tok.text for tok in NLP.tokenizer(sentence)]
    indexed = [text.vocab.stoi[t] for t in tokenized]
    tensor = torch.LongTensor(indexed).to(DEVICE)
    tensor = tensor.unsqueeze(1)
    tensor = torch.cat(2*[tensor], dim=1)
    preds = model(tensor)
    max_preds = preds.argmax(dim=1)
    return max_preds[0].item()

Simply, I just duplicated the row (acted as batch_size=2) to pass into the model, and extracted one of the result 😃 ~