如何在scikit学习中正确加载文本数据

Question

我按照这个例子为scikit-learn中的文本数据创建一个多项式朴素贝叶斯分类器。然而，混淆矩阵和分类器F-1分数的输出是不正确的。我认为这些错误与我使用的输入数据格式有关。每个培训示例我有一个csv文件。csv文件包含一行，其中包含诸如"废话、废话和soon"之类的功能。每个文件都被分类为正或负。我如何才能正确读取这些文件？

这是我的代码：

import numpy
import csv
from pandas import DataFrame
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import KFold
from sklearn.metrics import confusion_matrix, f1_score
NEWLINE = 'n'
NEGATIVE = 'negative'
POSITIVE = 'positive'
SOURCES = [
    ('negative\', NEGATIVE),
    ('positive\', POSITIVE)
]
SKIP_FILES = {'cmds'}

def build_data_frame(policies, path, classification):
    rows = []
    index = []
    for policy in policies:
        current_csv = path + policy + '.csv'
        # check if file exists
        if (os.path.isfile(current_csv)):
            with open(current_csv, 'r') as csvfile:
                reader = csv.reader(csvfile, delimiter=',', quotechar='"')
                # get each row in policy
                for row in reader:
                    # remove all commas from inside the text lists
                    clean_row = ' '.join(row)
                    rows.append({'text': clean_row, 'class': classification})
                    index.append(current_csv)
    data_frame = DataFrame(rows, index=index)
    return data_frame

def policy_analyzer_main(policies, write_pol_path):
    data = DataFrame({'text': [], 'class': []})
    for path, classification in SOURCES:
        data = data.append(build_data_frame(policies, write_pol_path + path, classification))
    classify(data)
pipeline = Pipeline([
    ('count_vectorizer',   CountVectorizer()),
    ('classifier',         MultinomialNB())
])
def classify(data):
    k_fold = KFold(n=len(data), n_folds=10)
    scores = []
    confusion = numpy.array([[0, 0], [0, 0]])
    for train_indices, test_indices in k_fold:
        train_text = data.iloc[train_indices]['text'].values
        train_y = data.iloc[train_indices]['class'].values.astype(str)
        test_text = data.iloc[test_indices]['text'].values
        test_y = data.iloc[test_indices]['class'].values.astype(str)
        pipeline.fit(train_text, train_y)
        predictions = pipeline.predict(test_text)
        confusion += confusion_matrix(test_y, predictions)
        score = f1_score(test_y, predictions, pos_label=POSITIVE)
        scores.append(score)
    print('Total emails classified:', len(data))
    print('Score:', sum(scores)/len(scores))
    print('Confusion matrix:')
    print(confusion)

以下是我收到的警告信息示例：

UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.
  'precision', 'predicted', average, warn_for)
 ('Total emails classified:', 75)
 ('Score:', 0.025000000000000001)
Confusion matrix:
[[39 35]
 [46 24]]

Answer 1

在每次迭代的训练测试拆分中查看您的predictions。因为该警告意味着，当测试集中的一些样本为阳性时，您的算法会将所有测试样本标记为阴性（可能只有1个样本为阳性，但无论如何，它都会发出该警告）。

还要查看数据集上的分割，因为有些测试分割可能只包含1个正样本，但分类器对其进行了错误分类

例如，在这种情况下，它会发出警告（为了弄清楚代码中发生了什么）：

from sklearn.metrics import f1_score
# here we have only 4 labels of 4 samples
f1_score([0,0,1,0],[0,0,0,0])
/usr/local/lib/python3.4/dist-packages/sklearn/metrics/classification.py:1074: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.
'precision', 'predicted', average, warn_for)