Google Cloud Functions支持含sklearn、pandas的Python吗？求部署资源指引

Absolutely! Google Cloud Functions does support running Python code with third-party packages like scikit-learn and pandas—you just need to handle dependencies correctly, which is the part that often trips people up when moving beyond basic "hello world" deployments. Let me walk you through exactly how to make this work:

You'll need two core files for your deployment:

• A main function file (e.g., main.py) with your code that uses pandas/sklearn
• A requirements.txt file listing all your dependencies

Example main.py #

import pandas as pd
from sklearn.linear_model import LinearRegression

def predict_value(request):
    # Example: Train a simple regression model and return a prediction
    # In a real use case, you'd pull input data from the request JSON
    training_data = pd.DataFrame({
        'feature1': [1, 2, 3, 4],
        'feature2': [10, 20, 30, 40]
    })
    training_target = [5, 10, 15, 20]
    
    model = LinearRegression()
    model.fit(training_data, training_target)
    
    # Predict for new input
    new_input = pd.DataFrame({'feature1': [5], 'feature2': [50]})
    prediction = model.predict(new_input)[0]
    
    return f"Predicted value: {round(prediction, 2)}"

Example requirements.txt #

Be specific with package versions to avoid compatibility issues with Google Cloud's Python runtime:

pandas==2.1.4
scikit-learn==1.3.2
numpy==1.24.3  # Scikit-learn depends on numpy, so explicit inclusion helps avoid conflicts

Assuming you've already set up the gcloud CLI, logged into your Google Cloud account, and set your active project, run this command from your function's directory:

gcloud functions deploy predict_value \
  --runtime python311 \
  --trigger-http \
  --allow-unauthenticated  # Optional: Remove this if you want to restrict access

• predict_value: The name of your function in main.py
• --runtime python311: Specifies the Python version (use a supported version like 3.9, 3.10, or 3.11)
• --trigger-http: Sets the function to be triggered via HTTP requests

Once deployment finishes, the gcloud CLI will output a URL for your function. You can test it directly in your browser or use curl:

curl https://YOUR_REGION-YOUR_PROJECT_ID.cloudfunctions.net/predict_value

You should see a response like Predicted value: 25.0.

• Deployment fails due to dependency conflicts: Double-check that your package versions are compatible with the Python runtime you're using. For example, older pandas versions may not support Python 3.11.
• Slow deployment: Packages like pandas and scikit-learn are large, so deployment may take a minute or two—this is normal.
• Missing system dependencies: Most common data science packages (including pandas/sklearn) work out of the box with Cloud Functions' pre-configured runtimes. If you hit issues with niche packages, you can use Cloud Functions Layers to package custom system dependencies.

内容的提问来源于stack exchange，提问作者Sean