RoBERTa 分类 RuntimeError: shape '[-1, 9]' 对于大小为 8 的输入无效。

Question

    m = MultiLabelBinarizer()

    X = pd.read_csv('data/data.csv', sep=None, engine='python')
    X = X.dropna()

    Y_train = m.fit_transform(X['labels'])
    Y_train2 = [list(i) for i in Y_train]

    data = pd.DataFrame({'text': pd.Series(X[text_col]), 'labels': Y_train2})
    data = data.dropna()

    train_df, eval_df = train_test_split(data, test_size=0.2)

    numLabels = len(pd.unique(X['labels])) # count of the labels

    model = MultiLabelClassificationModel('roberta', 'roberta-base', num_labels=numLabels, use_cuda=False)

    model.train_model(pd.DataFrame(train_df))

我的标签列的数据结构是：[[0,1,0,0,0,1,0,0],[0,1,1,0,0,1,0,0],[0,0,0,0,0,1]......]每一行都有一个标签列表，如标签列中的[0,1,0,0,0,1,0,0]。

而对于文字，每行有一篇文字（报纸文章）。

(从那个源头得到的。https:/github.comThilinaRajapaksesimpletransformers#minimal-start-for-multilabel-classification。)

如果我只用4个条目训练模型，模型可以被训练。但当我想用整个数据集来训练它时，它给我的答案是："RuntimeError: shape '[-1, 9]' is invalid input size 8::"。RuntimeError: shape '[-1, 9]' is invalid for input of size 8:

 File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/simpletransformers/classification/multi_label_classification_model.py", line 121, in train_model
    return super().train_model(train_df, multi_label=multi_label, eval_df=eval_df, output_dir=output_dir, show_running_loss=show_running_loss, args=args)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/simpletransformers/classification/classification_model.py", line 208, in train_model
    global_step, tr_loss = self.train(train_dataset, output_dir, multi_label=multi_label, show_running_loss=show_running_loss, eval_df=eval_df, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/simpletransformers/classification/classification_model.py", line 306, in train
    outputs = model(**inputs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/module.py", line 541, in __call__
    result = self.forward(*input, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/simpletransformers/custom_models/models.py", line 117, in forward
    loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1, self.num_labels))
RuntimeError: shape '[-1, 9]' is invalid for input of size 8

我不知道大小8是从哪里来的，现在该怎么办，因为它只对很少的条目起作用.有人能帮帮我吗？

Answer 1

[0,1,0,0,0,0,1,0,0] - 它是8个大小，但你的模型期望大小为9。这意味着，你的numLabels = 9。如果你有9个类，那么label-colum中的标签列表应该是这样的。[0,1,0,0,0,0,0,1,0,0,0].但我认为你只需要把num_labels传成8就可以了。