如何在Java Deeplearning4j中进行数据加载与预处理

<dependency>
    <groupId>org.deeplearning4j</groupId>
    <artifactId>deeplearning4j-core</artifactId>
    <version>1.0.0-M1.1</version> <!-- 请使用最新版本 -->
</dependency>
<dependency>
    <groupId>org.nd4j</groupId>
    <artifactId>nd4j-native-platform</artifactId>
    <version>1.0.0-M1.1</version>
</dependency>
<dependency>
    <groupId>org.datavec</groupId>
    <artifactId>datavec-api</artifactId>
    <version>1.0.0-M1.1</version>
</dependency>

import org.datavec.api.split.FileSplit;
import org.datavec.api.records.RecordReader;
import org.datavec.api.records.reader.impl.image.ImageRecordReader;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.factory.Nd4j;
import java.io.File;
import java.util.List;
public class ImageLoader {
    public static void main(String[] args) throws Exception {
        // 设置图像文件夹路径
        String imagePath = "path/to/image/folder";
        File file = new File(imagePath);
        // 创建 FileSplit
        FileSplit fileSplit = new FileSplit(file);
        // 创建 ImageRecordReader
        RecordReader recordReader = new ImageRecordReader(28, 28, 1); // 假设图像大小为28x28，单通道
        // 初始化读取器
        recordReader.initialize(fileSplit);
        // 读取图像
        INDArray image;
        while (recordReader.hasNext()) {
            image = recordReader.next();
            System.out.println(image.shapeInfoToString());
        }
    }
}

import org.datavec.api.split.FileSplit;
import org.datavec.api.records.reader.RecordReader;
import org.datavec.api.records.reader.impl.csv.CSVRecordReader;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.factory.Nd4j;
import java.io.File;
public class CSVLoader {
    public static void main(String[] args) throws Exception {
        // 设置 CSV 文件路径
        String csvFilePath = "path/to/data.csv";
        File file = new File(csvFilePath);
        // 创建 FileSplit
        FileSplit fileSplit = new FileSplit(file);
        // 创建 CSVRecordReader
        RecordReader recordReader = new CSVRecordReader();
        recordReader.initialize(fileSplit);
        // 读取数据
        INDArray data;
        while (recordReader.hasNext()) {
            List<String> record = recordReader.next();
            // 将数据转换为 INDArray
            double[] values = record.stream().mapToDouble(Double::parseDouble).toArray();
            data = Nd4j.create(values);
            System.out.println(data);
        }
    }
}

public static INDArray normalize(INDArray data) {
    double mean = data.meanNumber().doubleValue();
    double std = data.stdNumber().doubleValue();
    return data.sub(mean).div(std);
}

import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.dataset.api.MultiDataSet;
public class DataSplitter {
    public static void splitData(MultiDataSet dataSet) {
        // 假设数据集已加载为 MultiDataSet
        int trainSize = (int) (0.8 * dataSet.getFeatures(0).size(0)); // 80% 作为训练集
        INDArray trainData = dataSet.getFeatures(0).get(NDArrayIndex.interval(0, trainSize), NDArrayIndex.all());
        INDArray testData = dataSet.getFeatures(0).get(NDArrayIndex.interval(trainSize, dataSet.getFeatures(0).size(0)), NDArrayIndex.all());
        System.out.println("Training data shape: " + trainData.shapeInfoToString());
        System.out.println("Testing data shape: " + testData.shapeInfoToString());
    }
}