How to resolve Error in seqeval in NER bert finetuning?

问题描述 投票:0回答:1

我一直在尝试微调 NER 模型(bert/biobert),在第一个训练时期之后,在评估部分,我收到以下错误,知道哪里出了问题吗?

ValueError: Predictions and/or references don't match the expected format. Expected format: {'predictions': Sequence(feature=Value(dtype='string', id='label'), length=-1, id='sequence'),  'references': Sequence(feature=Value(dtype='string', id='label'), length=-1, id='sequence')}, Input predictions: [[2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  2, 2, 2, 2, 2, 2, 2, 2], [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], [2, 2,  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], ..., [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  2], [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]], Input references: [[2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], [2, 2, 2,  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], ..., [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2], [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2],  [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]]

我使用的是非常标准的 eval 函数,如果我从训练器中删除评估,模型训练没有任何问题并且结果很好,但我几乎没有指标。

    predictions, labels = p
    predictions = np.argmax(predictions, axis=2)

    true_predictions = [
        [label_list[p] for (p, l) in zip(prediction, label) if l != -100]
        for prediction, label in zip(predictions, labels)
    ]
    true_labels = [
        [label_list[l] for (p, l) in zip(prediction, label) if l != -100]
        for prediction, label in zip(predictions, labels)
    ]

    results = seqeval.compute(predictions=true_predictions, references=true_labels)
    return {
        "precision": results["overall_precision"],
        "recall": results["overall_recall"],
        "f1": results["overall_f1"],
        "accuracy": results["overall_accuracy"],
    }```


Thanks.
huggingface-transformers named-entity-recognition evaluation
1个回答
0
投票

如错误消息所示,预期的

predictions
references
应该是字符串列表而不是整数。对于
seqeval
,这是有道理的,因为
seqeval
指标与精确匹配的实体跨度有关(如标签的
B-
I-
前缀所示。

所以你的

label_list
应该是将标签标识符映射到标签标签,例如
["O", "B-PER", "I-PER", "B-ORG", "I-ORG", "B-LOC", "I-LOC"]
.

© www.soinside.com 2019 - 2024. All rights reserved.