如何在Android Studio中配置RapidMiner库并部署PMML预测模型

Hey there, let's tackle this PMML model integration issue for your Android object recognition app. I've been through similar hurdles, so here's what worked for me:

1. Ditch Direct RapidMiner Core Library Integration #

RapidMiner's core libraries are built for desktop environments, not mobile platforms. Trying to shoehorn them into Android Studio will only lead to dependency conflicts, bloated app size, and compatibility headaches. The rapidminer-extension-template you downloaded is meant for building desktop extensions for RapidMiner Studio— it won’t help your Android project, so don’t waste time on it.

2. Use an Android-Friendly PMML Parsing Library #

For Android, the most reliable lightweight PMML parser is the Android-compatible version of JPMML-Evaluator. Here’s how to set it up:

• Add these dependencies to your module-level build.gradle file:

  implementation 'org.jpmml:pmml-evaluator:1.6.4'
  implementation 'org.jpmml:pmml-evaluator-android:1.6.4'
  

  Make sure the versions match to avoid compatibility issues.

3. Load the PMML Model and Run Predictions #

First, move your RapidMiner-exported PMML file into your Android project’s assets directory. Then use this code to load the model and make predictions:

import org.jpmml.evaluator.Evaluator;
import org.jpmml.evaluator.EvaluatorUtil;
import org.jpmml.model.PMMLUtil;
import org.dmg.pmml.PMML;

import java.io.InputStream;
import java.util.HashMap;
import java.util.Map;

// Load your PMML model
private Evaluator loadPMMLModel() throws Exception {
    InputStream inputStream = getAssets().open("your_model.pmml");
    PMML pmml = PMMLUtil.unmarshal(inputStream);
    Evaluator evaluator = EvaluatorUtil.createEvaluator(pmml);
    evaluator.verify(); // Validate model integrity
    return evaluator;
}

// Run prediction with extracted features
private String predictObjectCategory(Evaluator evaluator, Map<String, Object> extractedFeatures) {
    // Map features to the model's required input format
    Map<String, ?> inputData = EvaluatorUtil.createArguments(evaluator, extractedFeatures);
    // Execute prediction
    Map<String, ?> predictionResult = evaluator.evaluate(inputData);
    // Retrieve the output (adjust the key to match your model's output field name)
    Object category = predictionResult.get("object_category");
    return category.toString();
}

4. Key Notes for Smooth Execution #

• Optimize the PMML Model: RapidMiner’s exported PMML often has redundant data. Use a PMML optimizer tool to trim it down— this will reduce load time and memory usage on Android.
• Run Predictions Off the UI Thread: Predictions can take time, so execute them in a background thread (like using Coroutines or AsyncTask) to avoid freezing your app’s UI.
• Validate Predictions: Test with sample feature sets to ensure the Android-side results match what you get in RapidMiner Studio. This helps catch issues with field mapping or PMML version mismatches.

内容的提问来源于stack exchange，提问作者John