报错TypeError：此数据类型不允许归约操作'argmax'，求代码问题排查

Hey there, let's break down this error and fix it step by step! That argmax error almost always boils down to non-numeric data types sneaking into your model inputs—either your target variable or your features aren't in a format scikit-learn can work with for classification tasks like Logistic Regression.

Why This Happens #

The argmax operation is used internally by scikit-learn to predict class labels (e.g., picking the highest probability class). It only works on numeric data types (int, float, etc.). If your target labels (y_train_data) are strings (like "positive"/"negative") or your features have non-numeric columns, this error pops up.

Step-by-Step Troubleshooting & Fixes #

1. Check Your Target Variable (y_train_data) #

First, verify the data type of your labels:

print(y_train_data.dtype)

If the output is object (meaning it's storing strings), you need to convert these to numeric labels using LabelEncoder:

from sklearn.preprocessing import LabelEncoder

# Convert string labels to numeric
le = LabelEncoder()
y_train_data = le.fit_transform(y_train_data)

This turns labels like "yes"/"no" into 1/0, which works with argmax.

2. Check Your Feature Data (x_train_data) #

Next, make sure all your features are numeric. Run this to see data types for each column:

print(x_train_data.dtypes)

If any columns show object, you'll need to encode them:

• For categorical features with no order (like "red"/"blue"/"green"), use one-hot encoding:

  x_train_data = pd.get_dummies(x_train_data)
  

• For ordered categorical features (like "low"/"medium"/"high"), use LabelEncoder instead.

3. Verify Your Incomplete DataFrame Code #

Your code cuts off at results_table = pd.DataFr...—if you're later using this DataFrame to store model results, make sure any columns used in argmax-related operations are numeric. Double-check that you're not accidentally storing strings or non-numeric values here.

Example Fixed Code #

Here's how your function might look with these fixes applied:

import pandas as pd
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.cross_validation import KFold, cross_val_score
from sklearn.metrics import confusion_matrix, precision_recall_curve, auc, roc_auc_score, roc_curve, recall_score, classification_report
from sklearn.preprocessing import LabelEncoder

def printing_Kfold_scores(x_train_data, y_train_data):
    # Fix 1: Convert target labels to numeric if needed
    if y_train_data.dtype == 'object':
        le = LabelEncoder()
        y_train_data = le.fit_transform(y_train_data)
    
    # Fix 2: Encode non-numeric features
    x_train_data = pd.get_dummies(x_train_data)
    
    fold = KFold(len(y_train_data), 5, shuffle=False)
    c_param_range = [0.01, 0.1, 1, 10, 100]
    results_table = pd.DataFrame(columns=['C_parameter', 'Mean recall score'])
    results_table['C_parameter'] = c_param_range
    
    # Cross-validation loop
    for i, c_param in enumerate(c_param_range):
        print(f'C parameter: {c_param}')
        recall_accs = []
        for iteration, indices in enumerate(fold, start=1):
            # Initialize model with current C parameter
            lr = LogisticRegression(C=c_param, penalty='l1')
            lr.fit(x_train_data.iloc[indices[0]], y_train_data.iloc[indices[0]])
            
            # Predict and calculate recall
            y_pred = lr.predict(x_train_data.iloc[indices[1]])
            recall_acc = recall_score(y_train_data.iloc[indices[1]], y_pred)
            recall_accs.append(recall_acc)
            print(f'Iteration {iteration}, recall score: {recall_acc}')
        
        # Store mean recall for this C parameter
        results_table.loc[i, 'Mean recall score'] = np.mean(recall_accs)
    
    # Optional: Print best C parameter
    best_c = results_table.loc[results_table['Mean recall score'].idxmax(), 'C_parameter']
    print(f'\nBest C parameter: {best_c}')
    return results_table

If the Error Persists #

If you still see the error, share the output of x_train_data.dtypes and y_train_data.dtype—that will help pinpoint exactly which column is causing the issue.

内容的提问来源于stack exchange，提问作者ROSA LILI