训练决策树回归模型时字符串转浮点数错误的解决方法

Hey there, I see you're hitting a common roadblock when working with categorical string features in scikit-learn models. That error pops up because tree-based models like DecisionTreeRegressor can only process numerical data—strings like "Nott'm Forest" can't be directly converted to floats. One-Hot Encoding is definitely the right approach here, so let's walk through how to implement it correctly to get your model training smoothly.

Step 1: Identify Categorical String Features #

First, let's confirm which columns are causing the issue. In football datasets, columns like HomeTeam, AwayTeam, or even Referee are typically categorical strings. You can quickly check this with:

import pandas as pd

# Load your dataset
df = pd.read_csv("your_football_data.csv")

# Print data types to spot string columns
print(df.dtypes)

# Extract list of categorical string columns
cat_string_cols = df.select_dtypes(include=['object']).columns.tolist()
print("Categorical string columns:", cat_string_cols)

Step 2: Handle Categorical Features with One-Hot Encoding #

There are two reliable ways to convert these strings to numerical data—pick the one that fits your workflow best:

Option 1: Quick Prototyping with pandas get_dummies #

This is perfect for small datasets or fast testing. It converts each unique category into a separate binary column:

# Separate features and target variable
X = df.drop("Full_Time_Home_Goals", axis=1)
y = df["Full_Time_Home_Goals"]

# Apply One-Hot Encoding to string columns
X_encoded = pd.get_dummies(X, columns=cat_string_cols, drop_first=True)

# Split into train/test sets
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X_encoded, y, test_size=0.2, random_state=42)

Option 2: Production-Ready with scikit-learn OneHotEncoder #

This method is better for avoiding data leakage (critical for robust models) and integrating into pipelines:

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder
from sklearn.model_selection import train_test_split

# Split data FIRST (never encode before splitting!)
X = df.drop("Full_Time_Home_Goals", axis=1)
y = df["Full_Time_Home_Goals"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Define preprocessor to encode categorical columns, keep numerical ones unchanged
preprocessor = ColumnTransformer(
    transformers=[
        ('cat_encoder', OneHotEncoder(sparse_output=False, drop='first'), cat_string_cols)
    ],
    remainder='passthrough'
)

# Fit encoder on training data only, then transform both sets
X_train_encoded = preprocessor.fit_transform(X_train)
X_test_encoded = preprocessor.transform(X_test)

Step 3: Train Your DecisionTreeRegressor #

Now that all features are numerical, you can train the model without errors:

from sklearn.tree import DecisionTreeRegressor

# Initialize and train the regressor
regressor = DecisionTreeRegressor(random_state=42)
regressor.fit(X_train_encoded, y_train)

# Make predictions on test data
y_pred = regressor.predict(X_test_encoded)

# Optional: Evaluate model performance
from sklearn.metrics import mean_absolute_error, mean_squared_error
print(f"Mean Absolute Error: {mean_absolute_error(y_test, y_pred)}")
print(f"Mean Squared Error: {mean_squared_error(y_test, y_pred)}")

Key Tips to Avoid Mistakes #

• Split before encoding: If you encode the full dataset first, you'll leak test set category info into training, which biases your model.
• Drop one category: Using drop_first=True removes redundant columns and avoids multicollinearity.
• Fix missing values first: Fill missing categorical values with a placeholder like "Unknown" or drop rows/columns before encoding.

内容的提问来源于stack exchange，提问作者PyNoob