利用huggingface进行文本分类

在 Hub 中，您可以找到 AI 社区共享的 27,000 多个模型，这些模型在情感分析、对象检测、文本生成、语音识别等任务上具有最先进的性能。

from transformers import pipeline
#sentiment_pipeline = pipeline("sentiment-analysis")
data = [
    "This is wonderful and easy to put together.  My cats love it.",
"This cat tree is almost perfect. I wanted a tall tree, and this one delivers. It reaches almost to the top of my 8\' ceiling",
"The super large box had disintegrated by the time it arrived to my doorstep & large portions were missing from a 89” solid wood cat tree. I took detailed pictures of the box before & after unpacking &  laying out all contents. Several pieces were badly damaged & 3 crucial pieces were missing.
A 45 minute phone call with Amazon resulted in Amazon requesting missing parts from Armarkat who never responded despite my repeated attempts to follow-through. Amazon offered for me to purchase another box, pack it & haul the box (weighs more than I weigh) to a place to be picked up. There’s no opportunity to do that where I live.

It’s a very expensive loss"]
sentiment_pipeline = pipeline("sentiment-analysis")

print(sentiment_pipeline(data))

在自己的亚马逊数据集上训练

 

 

from datasets import load_dataset
from transformers import AutoTokenizer
from transformers import DataCollatorWithPadding
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer
from datasets import load_dataset
import os.path as osp
import os
import numpy as np
from datasets import load_metric
### pretrained model :distilbert-base-uncased
### bert-base-uncased
### gpt2
### distilgpt2


def get_list(path,file_list):
    end_list = []
    for sample in file_list:
            sample_path = osp.join(path,sample)
            end_list.append(sample_path)
    return end_list

def get_dataset(dataset_path):
    test_path = osp.join(dataset_path,'test/')
    train_path = osp.join(dataset_path,'train/')
    val_path = osp.join(dataset_path,'val/')
    test_file_list = os.listdir(test_path)
    train_file_list = os.listdir(train_path)
    val_file_list = os.listdir(val_path)
    test_list = get_list(test_path,test_file_list)
    train_list = get_list(train_path,train_file_list)
    val_list = get_list(val_path,val_file_list)
    return test_list,train_list,val_list


def check_the_wrong_sample(labels,predictions):
    val_folder = '/cloud/cloud_disk/users/huh/dataset/nlp_dataset/question_dataset/process_data/cattree_product_quality/val'
    end_folder = '/cloud/cloud_disk/users/huh/dataset/nlp_dataset/question_dataset/process_data/cattree_product_quality/wrong_sample'
    sample_list = os.listdir(val_folder)
    index = 0
    
    for samle in labels:
        if samle != predictions[index]:
            print(index)
            print(sample_list[index])
            wrong_sample_path = osp.join(val_folder,sample_list[index])
            end_sample_path = osp.join(end_folder,sample_list[index])
            os.system("cp {} {}".format(wrong_sample_path,end_sample_path))
        index +=1
    



def compute_metric(eval_pred):
    metric = load_metric("accuracy")
    logits,labels = eval_pred
    print(logits,labels)
    print(len(logits),len(labels))
    predictions = np.argmax(logits,axis=-1)
    print(len(predictions))
    print('predictions')
    print(predictions)
    check_the_wrong_sample(labels,predictions)
    return metric.compute(predictions = predictions,references = labels)

def train(dataset_path):
    test_list,train_list,val_list = get_dataset(dataset_path)
    question_dataset = load_dataset('json', data_files={'train':train_list,'test':test_list,'val':val_list})
    tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
    def preprocess_function(examples):
        return tokenizer(examples["text"], truncation=True)
    tokenized_imdb = question_dataset.map(preprocess_function, batched=True)
    data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
    model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)

    training_args = TrainingArguments(
        output_dir="./results",
        learning_rate=2e-5,
        per_device_train_batch_size=16,
        per_device_eval_batch_size=16,
        num_train_epochs=5,
        weight_decay=0.01,
        logging_steps = 50,
        run_name = "catree",
        save_strategy='no'
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_imdb["train"],
        eval_dataset=tokenized_imdb["val"],
        tokenizer=tokenizer,
        data_collator=data_collator,
        compute_metrics=compute_metric
        
    )

    trainer.train()
    trainer.evaluate()



if __name__ == '__main__':
    #dataset_path = '/cloud/cloud_disk/users/huh/dataset/nlp_dataset/question_dataset/process_data/catree_personality_2.0'
    dataset_path = '/cloud/cloud_disk/users/huh/dataset/nlp_dataset/question_dataset/process_data/cattree_product_quality'
    train(dataset_path)