Python 中的逻辑回归：GridSearchCV 不起作用

Question

我对数据挖掘，尤其是文本分析有点陌生。我正在我的数据集上训练一个逻辑回归模型，并且我试图获得尽可能好的准确度，至少在 0.6 左右。但我似乎不能超过 0,5。 这是我的数据集：

df = pd.read_csv('https://raw.githubusercontent.com/cpedroni/DMML2021_Microsoft/main/data/training_data.csv')
df_pred = pd.read_csv('https://raw.githubusercontent.com/cpedroni/DMML2021_Microsoft/main/data/unlabelled_test_data.csv')

我使用带有 tfidf 的管道训练了我的模型：

tfidf_params = dict(sublinear_tf= True, 
                       min_df = 4, 
                       norm= 'l2', 
                       ngram_range= (1,4),
                       tokenizer= word_tokenize)

clf = Pipeline(steps=[
('features', TfidfVectorizer(**tfidf_params)),
('model', LogisticRegression(random_state=0, solver='lbfgs', max_iter=300))
])

clf.fit(X_train, y_train)

y_pred = clf.predict(X_test)
def metrics(y_test, y_pred):
  precision = precision_score(y_test, y_pred, average=None)
  recall = recall_score(y_test, y_pred, average=None)
  f1 = f1_score(y_test, y_pred, average=None)
  print("Precision: " + str(precision_score(y_test, y_pred, average='micro')))
  print("Recall: " + str(recall_score(y_test, y_pred, average='micro')))
  print("F1: " + str(2 * (precision * recall) / (precision + recall)))
  print("Accuracy: " + str(accuracy_score(y_test, y_pred)))

metrics(y_test, y_pred)

我得到了0.471875的准确度分数，但我想让它更高，所以尝试像这样进行网格搜索：

from sklearn.model_selection import GridSearchCV
param_grid_lr = {
    'max_iter': [20, 50, 100, 200, 500, 1000],                      
    'solver': ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga'],   
    'class_weight': ['balanced']                                    
}

logModel_grid = GridSearchCV(estimator=LogisticRegression(random_state=1234), param_grid=param_grid_lr, verbose=1, cv=10, n_jobs=-1)
logModel_grid.fit(X_train, y_train)
print(logModel_grid.best_estimator_)

但是我得到了这个我不明白的错误：ValueError: could not convert string to float。我收到此行logModel_grid.fit(X_train, y_train) 的此错误，但在进行网格搜索之前我没有收到日志模型的此错误。你知道为什么GridSearchCV 会带来这个错误吗？

Answer 1

您需要在估算器中包含矢量化器。

假设你是这样处理的：

from sklearn.model_selection import train_test_split, GridSearchCV
from nltk.tokenize import word_tokenize
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression

df = pd.read_csv('https://raw.githubusercontent.com/cpedroni/DMML2021_Microsoft/main/data/training_data.csv')
X_train, X_test, y_train, y_test = train_test_split(df['sentence'],df['difficulty'],
                                                    test_size=0.2, random_state=30, stratify=df['difficulty'])

我们像您的第一部分一样使用管道：

tfidf_params = dict(sublinear_tf= True, 
                       min_df = 4, 
                       norm= 'l2', 
                       ngram_range= (1,4),
                       tokenizer= word_tokenize)

pipe = Pipeline(steps=[
('features', TfidfVectorizer(**tfidf_params)),
('model', LogisticRegression())
])

定义参数，而不是下划线：

param_grid_lr = {
    'model__max_iter': [20, 50],                      
    'model__solver': ['newton-cg', 'lbfgs'],   
    'model__class_weight': ['balanced']                                    
}

适合：

logModel_grid = GridSearchCV(pipe, param_grid=param_grid_lr, verbose=1, cv=10, n_jobs=-1)
logModel_grid.fit(X_train, y_train)