sklearn.metrics.classification_report的support字段数值异常求助

Hey there! Let's break down why you're seeing that support mismatch in your classification_report output and how to fix it.

First, Identify the Root Cause #

Looking at your output, the avg / total row's support shows 18... — this is not a calculation error, but simply output truncation. Let's add up the support values for each class:

• Technology: 1, Travel:5, Fashion:25, Entertainment:130, Art:7, Politic:12
• Total: 1+5+25+130+7+12 = 180

The 18... is just your terminal cutting off the full number (180) because of limited width. The actual calculated support is correct!

Fix the Truncated Output #

Here are a few reliable ways to see the full, accurate report:

1. Capture the Report as a String First #

Instead of printing directly, save the report to a string and then print it — this avoids some terminal truncation issues:

from sklearn.metrics import classification_report
import numpy as np

# Your existing prediction code
y_pred = np.argmax(model.predict(X_test), axis=1)
y_true = np.argmax(y_test, axis=1)

# Capture full report as string
full_report = classification_report(y_true, y_pred, target_names=list(le.classes_))
print(full_report)

2. Use output_dict=True for Structured Results #

For precise access to all metrics (no printing issues), use the output_dict parameter to get a dictionary of results. You can then explicitly check the total support:

report_dict = classification_report(
    y_true, y_pred, 
    target_names=list(le.classes_),
    output_dict=True
)

# Check total support (matches sum of all class supports)
print(f"Total Support: {report_dict['weighted avg']['support']}")
# Note: In older sklearn versions, this might be under 'avg / total' instead of 'weighted avg'

3. Format as a DataFrame for Readability #

Using pandas to convert the report into a DataFrame ensures the full table is displayed without truncation:

import pandas as pd

report_df = pd.DataFrame(
    classification_report(
        y_true, y_pred, 
        target_names=list(le.classes_),
        output_dict=True
    )
).transpose()

print(report_df)

Bonus: Address the Class Imbalance Issue #

While not directly related to the support display problem, notice that most of your classes have 0 precision/recall/f1. This is because your dataset is heavily imbalanced (Entertainment has 130 samples, others have single-digit or low double-digit counts). The model is only learning to predict the dominant class, ignoring the rest. You might want to look into techniques like:

• Class weighting (class_weight='balanced' in most sklearn models)
• Oversampling minority classes (e.g., SMOTE)
• Undersampling the majority class

内容的提问来源于stack exchange，提问作者beginner