1 相关信息
2 引言
（1）机器学习方法在我们的任务中，没有重点研究，仅写出一个简单的baseline。该部分由队友博远实现，比赛最后的模型融合的部分，使用到，单个LGB模型，没有调参，5折交叉验证，线上都能达到0.79+的成绩，还有很多可提分点，如果时间允许，有很多提分技巧
- XGB模型
- LR模型
- 多模型融合EnsembleVoteClassifier
- 多模型融合StackingClassifier
- 加入伪标签训练
- 数据增强
（2）实现步骤 - 数据预处理
- K折划分数据
- TF-IDF提取特征，将单词转为数值矩阵
- 训练模型
- 每折预测一遍结果求和，最后取K折的结果的平均作为预测矩阵
- np.argmax取得预测值
- 生成提交文件
3 实现
import pandas as pd
from nltk.stem import WordNetLemmatizer
import re
import nltk
from spacy.lang.en.stop_words import STOP_WORDS
import numpy as np
import lightgbm as lgb
from sklearn import metrics
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.model_selection import KFold, StratifiedKFold
from sklearn.model_selection import train_test_split
clean_tag =True
if clean_tag ==True:
en_stop = set(nltk.corpus.stopwords.words('english'))
custom_stop_words = [
'doi', 'preprint', 'copyright', 'peer', 'reviewed', 'org', 'https', 'et', 'al', 'author', 'figure',
'rights', 'reserved', 'permission', 'used', 'using', 'biorxiv', 'medrxiv', 'license', 'fig', 'fig.',
'al.', 'elsevier', 'pmc', 'czi', 'www'
]
for word in custom_stop_words:
en_stop.add(word)
def preprocess_text(document):
stemmer = WordNetLemmatizer()
document = str(document)
document = document.replace("\n", ' ')
document = document.replace("/'", '')
document = re.sub(r'\W', ' ', document)
document = re.sub(r'\s+[a-zA-Z]\s+', ' ', document)
document = re.sub(r'\^[a-zA-Z]\s+', ' ', document)
document = re.sub(r'\s+', ' ', document, flags=re.I)
document = re.sub('[0-9]{5,}', '#####', document)
document = re.sub('[0-9]{4}', '####', document)
document = re.sub('[0-9]{3}', '###', document)
document = re.sub('[0-9]{2}', '##', document)
document = document.lower()
tokens = document.split()
tokens = [stemmer.lemmatize(word) for word in tokens]
tokens = [word for word in tokens if word not in en_stop]
tokens = [word for word in tokens if len(word) > 3]
preprocessed_text = ' '.join(tokens)
return preprocessed_text
train = pd.read_csv("train/train_stop.csv", sep="\t")
test = pd.read_csv("test/test.csv", sep="\t")
sub = pd.read_csv("sample_submit.csv")
train["text"] = train["title"] + " " + train["abstract"]
train["text"] = train["text"].progress_apply(lambda x: preprocess_text(x))
train.to_csv('ml_clean_data.csv', sep='\t')
else:
train = pd.read_csv('ml_clean_data.csv', sep='\t')
label_id2cate = dict(enumerate(train.categories.unique()))
label_cate2id = {value: key for key, value in label_id2cate.items()}
train["label"] = train["categories"].map(label_cate2id)
df = train[["text", "label"]]
df.head()
def submit_file(result_pred,label_id2cate):
print("存储预测结果")
sub=pd.read_csv('./sample_submit.csv')
sub['categories']=list(result_pred)
sub['categories']=sub['categories'].map(label_id2cate)
sub.to_csv('submit/submit_{}_ensemble.csv'.format(models_name), index=False)
params = {
"device_type": "gpu",
"max_depth": 5,
"min_data_in_leaf": 20,
"num_leaves": 35,
"learning_rate": 0.1,
"lambda_l1": 0.1,
"lambda_l2": 0.2,
"objective": "multiclass",
"num_class": 39,
"verbose": 0,
}
train_data = df["text"]
train_label = df["label"]
NFOLDS = 5
kfold = StratifiedKFold(n_splits=NFOLDS, shuffle=True, random_state=1)
kf = kfold.split(train_data, train_label)
cv_pred = np.zeros(test.shape[0])
valid_best = 0
for i, (train_fold, validate) in enumerate(kf):
X_train, X_validate, label_train, label_validate = (
train_data.iloc[train_fold],
train_data.iloc[validate],
train_label[train_fold],
train_label[validate],
)
vectorizer = CountVectorizer(max_features=50000)
tf_idf_transformer = TfidfTransformer()
tf_idf = tf_idf_transformer.fit_transform(vectorizer.fit_transform(X_train))
X_train_weight = tf_idf.toarray()
tf_idf = tf_idf_transformer.transform(vectorizer.transform(X_validate))
X_validate_weight = tf_idf.toarray()
dtrain = lgb.Dataset(X_train_weight, label_train)
dvalid = lgb.Dataset(X_validate_weight, label_validate, reference=dtrain)
bst = lgb.train(
params,
dtrain,
num_boost_round=10000,
valid_sets=dvalid,
early_stopping_rounds=500,
)
preds_last = bst.predict(test, num_iteration=bst.best_iteration)
cv_pred += bst.predict(test, num_iteration=bst.best_iteration)
valid_best += bst.best_score["valid_0"]["auc"]
cv_pred /= NFOLDS
valid_best /= NFOLDS
result =np.argmax(cv_pred,axis=1)
submit_file(list(result),label_id2cate)
|