python-3.x - 使用自定义数据集进行微调后，如何检查混淆矩阵？

Question

这个问题同How can I check a confusion_matrix after fine-tuning with custom datasets? ，在数据科学堆栈交换上。
背景
在使用自定义数据集进行微调后，我想检查一个混淆矩阵，包括精度、召回率和 f1 分数，如下所示。
微调过程和任务是Sequence Classification with IMDb Reviews在 Fine-tuning with custom datasets tutorial on Hugging face .
使用 Trainer 完成微调后，在这种情况下如何检查混淆矩阵？
一张confusion_matrix的图片，包括精度、召回率和f1-score original site : 只是例如输出图像

predictions = np.argmax(trainer.test(test_x), axis=1)

# Confusion matrix and classification report.
print(classification_report(test_y, predictions))

            precision    recall  f1-score   support

          0       0.75      0.79      0.77      1000
          1       0.81      0.87      0.84      1000
          2       0.63      0.61      0.62      1000
          3       0.55      0.47      0.50      1000
          4       0.66      0.66      0.66      1000
          5       0.62      0.64      0.63      1000
          6       0.74      0.83      0.78      1000
          7       0.80      0.74      0.77      1000
          8       0.85      0.81      0.83      1000
          9       0.79      0.80      0.80      1000

avg / total       0.72      0.72      0.72     10000

代码

from transformers import DistilBertForSequenceClassification, Trainer, TrainingArguments

training_args = TrainingArguments(
    output_dir='./results',          # output directory
    num_train_epochs=3,              # total number of training epochs
    per_device_train_batch_size=16,  # batch size per device during training
    per_device_eval_batch_size=64,   # batch size for evaluation
    warmup_steps=500,                # number of warmup steps for learning rate scheduler
    weight_decay=0.01,               # strength of weight decay
    logging_dir='./logs',            # directory for storing logs
    logging_steps=10,
)

model = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")

trainer = Trainer(
    model=model,                         # the instantiated 🤗 Transformers model to be trained
    args=training_args,                  # training arguments, defined above
    train_dataset=train_dataset,         # training dataset
    eval_dataset=val_dataset             # evaluation dataset
)

trainer.train()

到目前为止我做了什么
数据集准备Sequence Classification with IMDb Reviews ，我正在用 Trainer 进行微调。

from pathlib import Path

def read_imdb_split(split_dir):
    split_dir = Path(split_dir)
    texts = []
    labels = []
    for label_dir in ["pos", "neg"]:
        for text_file in (split_dir/label_dir).iterdir():
            texts.append(text_file.read_text())
            labels.append(0 if label_dir is "neg" else 1)

    return texts, labels

train_texts, train_labels = read_imdb_split('aclImdb/train')
test_texts, test_labels = read_imdb_split('aclImdb/test')

from sklearn.model_selection import train_test_split
train_texts, val_texts, train_labels, val_labels = train_test_split(train_texts, train_labels, test_size=.2)

from transformers import DistilBertTokenizerFast
tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-uncased')

train_encodings = tokenizer(train_texts, truncation=True, padding=True)
val_encodings = tokenizer(val_texts, truncation=True, padding=True)
test_encodings = tokenizer(test_texts, truncation=True, padding=True)

import torch

class IMDbDataset(torch.utils.data.Dataset):
    def __init__(self, encodings, labels):
        self.encodings = encodings
        self.labels = labels

    def __getitem__(self, idx):
        item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
        item['labels'] = torch.tensor(self.labels[idx])
        return item

    def __len__(self):
        return len(self.labels)

train_dataset = IMDbDataset(train_encodings, train_labels)
val_dataset = IMDbDataset(val_encodings, val_labels)
test_dataset = IMDbDataset(test_encodings, test_labels)

Answer 1

在这种情况下你可以做的是迭代验证集(或测试集)并手动创建 y_true 的列表。和 y_pred .

import torch
import torch.nn.functional as F
from sklearn import metrics
 
y_preds = []
y_trues = []
for index,val_text in enumerate(val_texts):
     tokenized_val_text = tokenizer([val_text], 
                                    truncation=True,
                                    padding=True,
                                    return_tensor='pt')
     logits = model(tokenized_val_text)
     prediction = F.softmax(logits, dim=1)
     y_pred = torch.argmax(prediction).numpy()
     y_true = val_labels[index]
     y_preds.append(y_pred)
     y_trues.append(y_true)

最后，

confusion_matrix = metrics.confusion_matrix(y_trues, y_preds, labels=["neg", "pos"]))
print(confusion_matrix)

观察:

模型的输出是 logits ，不是概率归一化。

因此，我们申请 softmax在第一维上转换为实际概率(例如 0.2% class 0 、 0.8% class 1 )。

我们申请 .argmax()操作以获取类的索引。