Python机器学习预测Y_Test时出现KeyError: action_taken求助

Hey there! Let's break down why you're hitting that frustrating KeyError and get your decision tree prediction workflow back on track.

The Root Cause of Your Error #

The issue boils down to a simple misstep in how you filtered your training data:

• You defined feature_variable as a list of columns to use as model features, but this list doesn't include your target column action_taken.
• Then you ran Training_logs = Training_logs[feature_variable] — this removes every column from your training DataFrame except the ones in feature_variable, including action_taken. Later when you try to access Training_logs['action_taken'], Python can't find it because you already deleted it!

Corrected Code Walkthrough #

Here's the fixed version of your code with key changes explained:

import pandas as pd #provide dataframe format
import numpy as np #support all high-level mathematical functions
from sklearn import tree #provides various ML feature such as various classification, regression and clustering algorithms
from sklearn.metrics import confusion_matrix
from sklearn.metrics import classification_report
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as plt #plotting library for python
import seaborn as sns #provide high level informative statistical and attractive graphics
from sklearn import metrics # for accuracy score
from sklearn.tree import export_graphviz
from sklearn.externals.six import StringIO
from IPython.display import Image
import pydotplus
from scipy import misc
%matplotlib inline

# Load and clean raw data
Training_logs=pd.read_csv('C:/Users/path/PA_training_data_reformed_2.CSV')
Training_logs=Training_logs.fillna('')
Testing_logs=pd.read_csv('C:/Users/path/PA_testing_data_reformed_2.csv')
Testing_logs= Testing_logs.fillna('')

# Define categorical value lists (unchanged)
corrected_data=['deny','drop','reset-client','reset-server','reset-both','block-url','block-ip','random-drop','sinkhole','syncookie-sent','block-continue','continue','block-overide','override-lockout','override']
severity= ['medium','high','critical']
action=['alert','allow','corrected_data']
event_category=['THREAT','CORRELATION']
sub_category=['spyware','url','virus','vulnerability','wildfire','wildfire-virus']
traffic_direction=['NS','EW','SN']

# Define feature columns (keep target column OUT of this list!)
feature_variable=['severity','event_category', 'action', 'sub_category','traffic_direction']

# --------------------------
# Key Fix 1: Separate features and target BEFORE filtering
# --------------------------
# Extract features from training data (keep target column intact in original df)
X_train_raw = Training_logs[feature_variable]
# Extract target variable directly from original training data
y_train_raw = Training_logs['action_taken']

# Process features with DictVectorizer
from sklearn.feature_extraction import DictVectorizer
from sklearn.preprocessing import LabelEncoder

X_dict = X_train_raw.to_dict(orient='records')
vect = DictVectorizer(sparse=False)
X_vector = vect.fit_transform(X_dict)

# Encode target variable
le = LabelEncoder()
y_train = le.fit_transform(y_train_raw[:-1])
X_train = X_vector[:-1]

# --------------------------
# Key Fix 2: Correct test data processing
# Use the fitted vectorizer from training to transform test data
# --------------------------
X_test_raw = Testing_logs[feature_variable]
X_test_dict = X_test_raw.to_dict(orient='records')
X_test = vect.transform(X_test_dict)

# Now train your model and make predictions
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)

# Optional: If your test data has action_taken for evaluation, uncomment below
# y_test = le.transform(Testing_logs['action_taken'])
# print(classification_report(y_test, y_pred))
# print(confusion_matrix(y_test, y_pred))

Quick Best Practices to Avoid This in the Future #

• Always separate features and target explicitly: Never filter your entire dataset down to just features before extracting your target column.
• Reuse fitted transformers on test data: Always use the vectorizer/encoder you fitted on training data to process test data — don't refit or reuse training data vectors for testing.
• Double-check column existence: Before accessing any column, you can run print(Training_logs.columns) to confirm it's still present in your DataFrame.

内容的提问来源于stack exchange，提问作者Murtuza Husain