我需要通过公共行或列标签合并两个大的csv文件(每个so~500mb中大约有4000万个数据元素)，这可以由用户指定。例如，如果数据集1.csv包含：
patient_id x1 x2 x3
pi1 1 2 3
pi3 4 5 6
包含数据集2.csv：
patient_id y1 y2 y3
pi0 0 0 0
pi1 11 12 13
pi2 99 98 97
pi3 14 15 16
用户可以指定通过行标签(患者ID)合并这两个文件，结果输出.csv为：
patient_id x1 x2 x3 y1 y2 y3
pi1 1 2 3 11 12 13
pi3 4 5 6 14 15 16
因为我们只组合了两个输入文件共同的(交集)患者id的信息。我解决这个问题的策略是创建一个HashMap，其中要合并的行或列标签(在本例中是行标签，即患者id)是键，患者id的数据作为值存储在ArrayList中。我为每个输入数据文件创建一个HashMap，然后根据类似的键合并值。我将数据表示为ArrayList>类型的二维ArrayList，因此合并的数据也具有此类型。然后，我只需遍历合并的ArrayList>对象(我称之为Data类型对象)，并将其打印到文件中。代码如下：
以下是依赖于以下数据类文件的DataMerge类。
import java.util.HashMap;
import java.util.ArrayList;
public class DataMerge {
/**Merges two Data objects by a similar label. For example, if two data sets represent
* different data for the same set of patients, which are represented by their unique patient
* ID, mergeData will return a data set containing only those patient IDs that are common to both
* data sets along with the data represented in both data sets. labelInRow1 and labelInRow2 separately
* indicate whether the common labels are in separate rows(true) of d1 and d2, respectively, or separate columns otherwise.*/
public static Data mergeData(Data d1, Data d2, boolean labelInRow1,
boolean labelInRow2){
ArrayList<ArrayList<String>> mergedData = new ArrayList<ArrayList<String>>();
HashMap<String,ArrayList<String>> d1Map = d1.mapFeatureToData(labelInRow1);
HashMap<String,ArrayList<String>> d2Map = d2.mapFeatureToData(labelInRow2);
ArrayList<String> d1Features;
ArrayList<String> d2Features;
if (labelInRow1){
d1Features = d1.getColumnLabels();
} else {
d1Features = d1.getRowLabels();
}
if (labelInRow2){
d2Features = d2.getColumnLabels();
} else {
d2Features = d2.getRowLabels();
}
d1Features.trimToSize();
d2Features.trimToSize();
ArrayList<String> mergedFeatures = new ArrayList<String>();
if ((d1.getLabelLabel() != "") && (d1.getLabelLabel() == "")) {
mergedFeatures.add(d1.getLabelLabel());
}
else if ((d1.getLabelLabel() == "") && (d1.getLabelLabel() != "")) {
mergedFeatures.add(d2.getLabelLabel());
} else {
mergedFeatures.add(d1.getLabelLabel());
}
mergedFeatures.addAll(d1Features);
mergedFeatures.addAll(d2Features);
mergedFeatures.trimToSize();
mergedData.add(mergedFeatures);
for (String key : d1Map.keySet()){
ArrayList<String> curRow = new ArrayList<String>();
if (d2Map.containsKey(key)){
curRow.add(key);
curRow.addAll(d1Map.get(key));
curRow.addAll(d2Map.get(key));
curRow.trimToSize();
mergedData.add(curRow);
}
}
mergedData.trimToSize();
Data result = new Data(mergedData, true);
return result;
}
}
下面是Data类型对象及其关联的HashMap生成函数，以及一些行和列标签提取方法。
import java.util.*;
import java.io.*;
/**Represents an unlabeled or labeled data set as a series of nested ArrayLists, where each nested
* ArrayList represents a line of the input data.*/
public class Data {
private ArrayList<String> colLabels = new ArrayList<String>(); //row labels
private ArrayList<String> rowLabels = new ArrayList<String>(); //column labels
private String labelLabel;
private ArrayList<ArrayList<String>> unlabeledData; //data without row and column labels
/**Returns an ArrayList of ArrayLists, where each nested ArrayList represents a line
*of the input file.*/
@SuppressWarnings("resource")
private static ArrayList<ArrayList<String>> readFile(String filePath, String fileSep){
ArrayList<ArrayList<String>> result = new ArrayList<ArrayList<String>>();
try{
BufferedReader input = new BufferedReader(new FileReader (filePath));
String line = input.readLine();
while (line != null){
String[] splitLine = line.split(fileSep);
result.add(new ArrayList<String>(Arrays.asList(splitLine)));
line = input.readLine();
}
}
catch (Exception e){
System.err.println(e);
}
result.trimToSize();;
return result;
}
/**Returns an ArrayList of ArrayLists, where each nested ArrayList represents a line of the input
* data but WITHOUT any row or column labels*/
private ArrayList<ArrayList<String>> extractLabelsAndData(String filePath, String fileSep){
ArrayList<ArrayList<String>> tempData = new ArrayList<ArrayList<String>>();
tempData.addAll(readFile(filePath, fileSep));
tempData.trimToSize();
this.colLabels.addAll(tempData.remove(0));
this.labelLabel = this.colLabels.remove(0);
this.colLabels.trimToSize();
for (ArrayList<String> line : tempData){
this.rowLabels.add(line.remove(0));
}
this.rowLabels.trimToSize();
return tempData;
}
/**Returns an ArrayList of ArrayLists, where each nested ArrayList represents a line of the input
* data but WITHOUT any row or column labels. Does mutate the original data*/
private ArrayList<ArrayList<String>> extractLabelsAndData (ArrayList<ArrayList<String>> data){
ArrayList<ArrayList<String>> result = new ArrayList<ArrayList<String>>();
for (ArrayList<String> line : data){
ArrayList<String> temp = new ArrayList<String>();
for (String element : line){
temp.add(element);
}
temp.trimToSize();
result.add(temp);
}
this.colLabels.addAll(result.remove(0));
this.labelLabel = this.colLabels.remove(0);
this.colLabels.trimToSize();
for (ArrayList<String> line : result){
this.rowLabels.add(line.remove(0));
}
this.rowLabels.trimToSize();
result.trimToSize();
return result;
}
/**Returns the labelLabel for the data*/
public String getLabelLabel(){
return this.labelLabel;
}
/**Returns an ArrayList of the labels while maintaining the order
* in which they appear in the data. Row indicates that the desired
* features are all in the same row. Assumed that the labels are in the
* first row of the data. */
public ArrayList<String> getColumnLabels(){
return this.colLabels;
}
/**Returns an ArrayList of the labels while maintaining the order
* in which they appear in the data. Column indicates that the desired
* features are all in the same column. Assumed that the labels are in the
* first column of the data.*/
public ArrayList<String> getRowLabels(){
return this.rowLabels;
}
/**Creates a HashMap where a list of feature labels are mapped to the entire data. For example,
* if a data set contains patient IDs and test results, this function can be used to create
* a HashMap where the keys are the patient IDs and the values are an ArrayList of the test
* results. The boolean input isRow, which, when true, designates that the
* desired keys are listed in the rows or false if they are in the columns.*/
public HashMap<String, ArrayList<String>> mapFeatureToData(boolean isRow){
HashMap<String, ArrayList<String>> featureMap = new HashMap<String,ArrayList<String>>();
if (!isRow){
for (ArrayList<String> line : this.unlabeledData){
for (int i = 0; i < this.colLabels.size(); i++){
if (featureMap.containsKey(this.colLabels.get(i))){
featureMap.get(this.colLabels.get(i)).add(line.get(i));
} else{
ArrayList<String> firstValue = new ArrayList<String>();
firstValue.add(line.get(i));
featureMap.put(this.colLabels.get(i), firstValue);
}
}
}
} else {
for (int i = 0; i < this.rowLabels.size(); i++){
if (!featureMap.containsKey(this.rowLabels.get(i))){
featureMap.put(this.rowLabels.get(i), this.unlabeledData.get(i));
} else {
featureMap.get(this.rowLabels.get(i)).addAll(this.unlabeledData.get(i));
}
}
}
return featureMap;
}
/**Writes the data to a file in the specified outputPath. sep indicates the data delimiter.
* labeledOutput indicates whether or not the user wants the data written to a file to be
* labeled or unlabeled. If the data was unlabeled to begin with, then labeledOutput
* should not be set to true. */
public void writeDataToFile(String outputPath, String sep){
try {
PrintStream writer = new PrintStream(new BufferedOutputStream (new FileOutputStream (outputPath, true)));
String sol = this.labelLabel + sep;
for (int n = 0; n < this.colLabels.size(); n++){
if (n == this.colLabels.size()-1){
sol += this.colLabels.get(n) + "n";
} else {
sol += this.colLabels.get(n) + sep;
}
}
for (int i = 0; i < this.unlabeledData.size(); i++){
ArrayList<String> line = this.unlabeledData.get(i);
sol += this.rowLabels.get(i) + sep;
for (int j = 0; j < line.size(); j++){
if (j == line.size()-1){
sol += line.get(j);
} else {
sol += line.get(j) + sep;
}
}
sol += "n";
}
sol = sol.trim();
writer.print(sol);
writer.close();
} catch (Exception e){
System.err.println(e);
}
}
/**Constructor for Data object. filePath specifies the input file directory,
* fileSep indicates the file separator used in the input file, and hasLabels
* designates whether the input data has row and column labels. Note that if
* hasLabels is set to true, it is assumed that there are BOTH row and column labels*/
public Data(String filePath, String fileSep, boolean hasLabels){
if (hasLabels){
this.unlabeledData = extractLabelsAndData(filePath, fileSep);
this.unlabeledData.trimToSize();
} else {
this.unlabeledData = readFile(filePath, fileSep);
this.unlabeledData.trimToSize();
}
}
/**Constructor for Data object that accepts nested ArrayLists as inputs*/
public Data (ArrayList<ArrayList<String>> data, boolean hasLabels){
if (hasLabels){
this.unlabeledData = extractLabelsAndData(data);
this.unlabeledData.trimToSize();
} else {
this.unlabeledData = data;
this.unlabeledData.trimToSize();
}
}
}
该程序适用于小型数据集，但已经运行了5天多，合并仍未完成。我正在寻找一个更有效的时间和内存解决方案。有人建议使用字节数组而不是字符串，这可能会使它运行得更快。有人有什么建议吗？
EDIT：我在代码中做了一些挖掘，发现读取输入文件并合并它们几乎不需要时间(比如20秒)。写入文件需要5天以上的时间
您将数百万行数据的所有数据字段连接到一个巨大的字符串中，然后写入该字符串。这是由于在分配和重新分配超大字符串时内存抖动造成的缓慢死亡，将它们一遍又一遍地复制到要添加到字符串中的每个字段和分隔符中。大约在第三天或第四天，每条字符串都是。。。数百万个字符长。。。而你那可怜的垃圾收集器却汗流浃背。
不要那样做。
分别构建输出文件的每一行并编写它。然后构建下一行。
此外，使用StringBuilder类来构建行，尽管您在前一步中会得到这样的改进，但您可能根本不需要为此而烦恼。尽管这是一种方法，你应该学会如何做。