段落向量模型的交叉验证

Question

我刚刚在尝试对段落向量模型应用交叉验证时遇到错误：

import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
from sklearn.pipeline import Pipeline
from gensim.sklearn_api import D2VTransformer

data = pd.read_csv('https://pastebin.com/raw/bSGWiBfs')
np.random.seed(0)

X_train = data.apply(lambda r: simple_preprocess(r['text'], min_len=2), axis=1)
y_train = data.label

model = D2VTransformer(size=10, min_count=1, iter=5, seed=1)
clf = LogisticRegression(random_state=0)

pipeline = Pipeline([
        ('vec', model),
        ('clf', clf)
    ])

pipeline.fit(X_train, y_train)

score = pipeline.score(X_train, y_train)
print("Score:", score) # This works
cval = cross_val_score(pipeline, X_train, y_train, scoring='accuracy', cv=3)
print("Cross-Validation:", cval) # This doesn't work

KeyError: 0

我尝试用 model.transform(X_train) 或 model.fit_transform(X_train) 替换 cross_val_score 中的 X_train。此外，我对原始输入数据 (data.text) 进行了同样的尝试，而不是预处理文本。我怀疑与 Pipeline 的 .score 函数相比，交叉验证的 X_train 格式一定有问题，后者工作得很好。我还注意到 cross_val_score 与 CountVectorizer() 一起工作。

有人发现错误了吗？

Answer 1

不，这与model的转换无关。它与 cross_val_score.

有关

cross_val_score 将根据 cv 参数拆分提供的数据。为此，它会做这样的事情：

for train, test in splitter.split(X_train, y_train):
    new_X_train, new_y_train = X_train[train], y_train[train]

但是您的 X_train 是一个 pandas.Series 对象，其中基于索引的选择不能像这样工作。看这个：https://pandas.pydata.org/pandas-docs/stable/indexing.html#selection-by-position

更改此行：

X_train = data.apply(lambda r: simple_preprocess(r['text'], min_len=2), axis=1)

至：

# Access the internal numpy array
X_train = data.apply(lambda r: simple_preprocess(r['text'], min_len=2), axis=1).values

OR

# Convert series to list
X_train = data.apply(lambda r: simple_preprocess(r['text'], min_len=2), axis=1).tolist()

段落向量模型的交叉验证

Cross-validation for paragraph-vector model

scikit-learn

transform

cross-validation

gensim