/*
* Run all routine to execute the training
*/
package biodiscml;
import java.io.BufferedReader;
import java.io.File;
import java.io.FileReader;
import java.io.FileWriter;
import java.io.IOException;
import java.io.InputStreamReader;
import java.io.PrintWriter;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.ArrayList;
import java.util.HashMap;
import java.nio.file.StandardCopyOption;
/**
*
* @author Mickael
*/
public class Main {
public static boolean debug = false;
public static boolean debug2 = false;
public static boolean printFailedModels = false; //print errors of failed models
public static boolean isClassification = true;
public static String wd = "";
public static String project = "myProject";
//program functions
public static String configFile = ""; //config file
public static boolean needConfigFile = true;
public static boolean training = false;
public static boolean predictNewData = false;
public static boolean trainingBestModel = false;
public static HashMap<String, String> hmTrainingBestModelList = new HashMap<>();//modelID, identifier prefix
public static String modelFile = "";
public static HashMap<String, String> hmExcludedFeatures = new HashMap<>();//features to exclude from the final dataset
//config
public static String mergingID = "Instance";
//source files
public static HashMap<String, String> hmTrainFiles = new HashMap<>();//filename, identifier prefix
public static HashMap<String, String> hmNewDataFiles = new HashMap<>();//filename, identifier prefix
//options
public static Boolean doClassification = false;
public static String classificationClassName = "class";
public static String regressionClassName = "class";
public static String separator = "";
public static Boolean doRegression = false;
public static Boolean classificationFastWay = false;
public static Integer numberOfBestModels = 1;
public static ArrayList<String> classificationFastWayCommands = new ArrayList<>(); //classifier, optimizer
public static ArrayList<String> classificationBruteForceCommands = new ArrayList<>(); //classifier, optimizer
public static String classificationOptimizers = "auc, mcc, fdr, ber, acc";
public static String searchmodes = "f,fb,b,bf,top1,top5,top10,top15,top20,top30,top40,top50,top75,top100,top200,all";
public static Boolean metaCostSensitiveClassifier = false;
public static Boolean regressionFastWay = false;
public static String regressionOptimizers = "CC, RMSE";
public static Boolean metaAdditiveRegression = false;
public static ArrayList<String> regressionFastWayCommands = new ArrayList<>(); //classifier, optimizer
public static ArrayList<String> regressionBruteForceCommands = new ArrayList<>(); //classifier, optimizer
public static double pAUC_lower = 0;
public static double pAUC_upper = 0.3;
public static double spearmanCorrelation_lower = -0.99;
public static double spearmanCorrelation_upper = 0.99;
public static double pearsonCorrelation_lower = -0.99;
public static double pearsonCorrelation_upper = 0.99;
public static String bestModelsSortingMetric = "AVG_MCC";
public static double bestModelsSortingMetricThreshold = 0.1;
public static Integer maxNumberOfFeaturesInModel = 200;
public static int maxNumberOfSelectedFeatures = 1000;
public static boolean doSampling = true;
public static int samplingFold = 3; //separate the set in x parts, keep 1 for test, others for training
public static int bootstrapAndRepeatedHoldoutFolds = 100; // Also used for repeated holdout
public static String cpus = "max";
public static boolean combineModels = false;
public static boolean retrieveCorrelatedGenes = true;
public static String combinationRule = "AVG";
public static double maxRankingScoreDifference = 0.005; //for correlated gene retreiving
public static boolean loocv = true;
public static boolean repeatedHoldout = true;
public static boolean bootstrap = true;
public static boolean computeBestModel = true;
public static boolean resumeTraining = false;
public static boolean restoreRun = false;
public static String previousRunPath = "";
public static String previousRunProjectName = "";
public static Boolean noFeatureSelection = false;
public static boolean generateModelWithCorrelatedGenes = false;
public static String missingValueToReplace = "?"; //for testing, if a feature is missing, set a value to replace missing data
//benchmark
public static String bench_AUC = "";
//TODO
static boolean retreiveCorrelatedGenesByRankingScore = true; //avoid for non-binary classes and regression
static boolean ROCcurves = false; //experimental
static boolean UpSetR = false; //experimental
static boolean performShortTest = true;
public static void main(String[] args) throws IOException {
System.out.println("#### BioDiscML ####\n");
// check java version
String version = System.getProperty("java.version");
if (!version.contains("1.8")) {
String arg = "";
for (String argl : args) {
arg += argl + " ";
}
System.out.println("ERROR: Java version is " + version + "."
+ "\nYou need java 1.8. If it is installed, try this command:\n"
);
if (System.getProperty("os.name").toLowerCase().contains("windows")) {
System.out.println("\"c:\\Program Files\\Java\\jdk1.8.0_111\\bin\\java.exe\" -jar biodiscml.jar " + arg);
} else {
System.out.println("/mnt/software/jvm/jdk1.8.0_371/bin/java -jar biodiscml.jar " + arg);
}
System.exit(0);
}
//read configuration file
System.out.println("#### Parsing options...");
setOptionsFromCommandLine(args); //from command line
if (!configFile.isEmpty() && needConfigFile) {
setConfiguration();
}
//set models
if (!classificationFastWay && !regressionFastWay) {
try {
String line = "";
File classifiers = new File(wd + "/" + "classifiers.conf");
BufferedReader br;
if (!classifiers.exists()) {
try {
br = new BufferedReader(new InputStreamReader(Main.class.getResourceAsStream("/classifiers.conf")));
} catch (Exception e) {
br = new BufferedReader(new FileReader("/classifiers.conf"));
}
} else {
br = new BufferedReader(new FileReader(classifiers));
}
line = "";
while (br.ready()) {
if (!line.startsWith("#") && !line.trim().isEmpty()) {
String option = line.split("=")[0].trim();
String value = line.split("=")[1].trim();
switch (option) {
case "ccmd":
classificationBruteForceCommands.add(value.trim());
break;
case "rcmd":
regressionBruteForceCommands.add(value.trim());
break;
}
}
line = br.readLine();
}
} catch (Exception e) {
e.printStackTrace();
}
} else {
System.out.println("Model search mode: Fast way mode");
}
//set number of max cpus to use
if (!cpus.equals("max")) {
System.setProperty("java.util.concurrent.ForkJoinPool.common.parallelism", cpus);
}
// Retrieve previously existing a (train and test data) and b (infogain) files
if (restoreRun) {
resumeTraining = true;
// get current directory
if (wd.isEmpty()) {
wd = new java.io.File(".").getCanonicalPath() + java.io.File.separator;
}
System.out.println("Restoring data from previous project " + previousRunProjectName
+ "\nCopying files from " + previousRunPath);
//copy previous run
try {
Path source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.all_data.arff");
Path destination = Paths.get(wd + "/" + project + "a.classification.all_data.arff");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.all_data.csv");
destination = Paths.get(wd + "/" + project + "a.classification.all_data.csv");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.data_to_test.arff");
destination = Paths.get(wd + "/" + project + "a.classification.data_to_test.arff");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.data_to_test.csv");
destination = Paths.get(wd + "/" + project + "a.classification.data_to_test.csv");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.data_to_train.arff");
destination = Paths.get(wd + "/" + project + "a.classification.data_to_train.arff");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_a.classification.data_to_train.csv");
destination = Paths.get(wd + "/" + project + "a.classification.data_to_train.csv");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_b.featureSelection.infoGain.arff");
destination = Paths.get(wd + "/" + project + "b.featureSelection.infoGain.arff");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
source = Paths.get(previousRunPath + previousRunProjectName + "_b.featureSelection.infoGain.csv");
destination = Paths.get(wd + "/" + project + "b.featureSelection.infoGain.csv");
Files.copy(source, destination, StandardCopyOption.REPLACE_EXISTING);
File f = new File(previousRunPath + previousRunProjectName + "_c.classification.results.csv");
f.createNewFile();
} catch (Exception e) {
e.printStackTrace();
}
}
// Go to training
if (training) {
System.out.println("#### Start training...");
//CLASSIFICATION
if (doClassification) {
isClassification = true;
//put data together in the same file for ML
System.out.println("## Preprocessing of the input file(s)");
String CLASSIFICATION_FILE = wd + project + "a.classification.data_to_train.csv"; //output of AdaptDatasetToWeka()
if (debug) {
System.out.println("CLASSIFICATION_FILE: " + CLASSIFICATION_FILE);
}
if (new File(CLASSIFICATION_FILE).exists() && resumeTraining) {
System.out.println("Preprocessing of the input file(s) already done... skipping");
} else {
AdaptDatasetToTraining c = new AdaptDatasetToTraining(CLASSIFICATION_FILE);
}
//execute feature selection and training
System.out.println("## Feature selection and training");
String FEATURE_SELECTION_FILE = wd + project + "b.featureSelection.infoGain.csv"; // output of Training(), feature selection result
String TRAINING_RESULTS_FILE = wd + project + "c.classification.results.csv"; // output of Training(), models performances
Training m = new Training(CLASSIFICATION_FILE, TRAINING_RESULTS_FILE, FEATURE_SELECTION_FILE, "class");
//choose best model
if (computeBestModel) {
System.out.println("## Best model selection");
BestModelSelectionAndReport b = new BestModelSelectionAndReport(CLASSIFICATION_FILE, FEATURE_SELECTION_FILE, TRAINING_RESULTS_FILE,
"classification");
}
}
//REGRESSION
if (doRegression) {
isClassification = false;
//put data together in the same file for ML
System.out.println("## Preprocessing of the input file(s)");
String REGRESSION_FILE = wd + project + "a.regression.data_to_train.csv";
if (new File(REGRESSION_FILE).exists() && resumeTraining) {
System.out.println("Preprocessing of the input file(s) already done... skipped by resumeTraining");
} else {
AdaptDatasetToTraining c = new AdaptDatasetToTraining(REGRESSION_FILE);
}
//execute training
System.out.println("## Feature selection and training");
String FEATURE_SELECTION_FILE = wd + project + "b.featureSelection.RELIEFF.csv";//filled by feature selection algo type
String TRAINING_RESULTS_FILE = wd + project + "c.regression.results.csv";
Training m = new Training(REGRESSION_FILE, TRAINING_RESULTS_FILE, FEATURE_SELECTION_FILE, "reg");
//choose best model
if (computeBestModel) {
System.out.println("## Best model selection");
BestModelSelectionAndReport b = new BestModelSelectionAndReport(REGRESSION_FILE, FEATURE_SELECTION_FILE, TRAINING_RESULTS_FILE,
"regression");
}
}
if (!doClassification && !doRegression) {
System.err.println("[error] No prediction type has been set (classification or regression)."
+ " Set doClassification or doRegression at true");
System.exit(0);
}
}
if (predictNewData) {
System.out.println("#### Start predicting new data...");
if (modelFile.isEmpty()) {
System.err.println("[error] No model file have been provided (Set a modelFile in config file)");
System.exit(0);
}
if (hmNewDataFiles.isEmpty()) {
System.err.println("[error] No new data file have been provided (Set a newDataFile in config file)");
System.exit(0);
}
//put data together in the same file for ML
String NEWDATA_FILE = wd + project + ".data_to_predict.csv"; //output of AdaptDatasetToWeka()
AdaptDatasetToTesting c = null;
if (doClassification) {
c = new AdaptDatasetToTesting(classificationClassName, hmNewDataFiles,
NEWDATA_FILE, separator, wd + modelFile);
} else {
c = new AdaptDatasetToTesting(regressionClassName, hmNewDataFiles,
NEWDATA_FILE, separator, wd + modelFile);
}
//execute feature selection and training
String PREDICTIONS_RESULTS_FILE = wd + project + modelFile + "_" + ".prediction.results.txt"; // output of Testing
TestingAndEvaluate t = new TestingAndEvaluate();
t.TestingAndEvaluate(wd + modelFile, NEWDATA_FILE, PREDICTIONS_RESULTS_FILE, c.isMissingClass());
}
if (trainingBestModel) {
System.out.println("#### Start best model selection...");
if (doClassification) {
String CLASSIFICATION_FILE = wd + project + "a.classification.data_to_train.csv";
String TRAINING_RESULTS_FILE = wd + project + "c.classification.results.csv"; // output of Training(), models performances
String FEATURE_SELECTION_FILE = wd + project + "b.featureSelection.infoGain.csv"; // output of Training(), feature selection result
BestModelSelectionAndReport b = new BestModelSelectionAndReport(CLASSIFICATION_FILE, FEATURE_SELECTION_FILE, TRAINING_RESULTS_FILE,
"classification");
} else {
String REGRESSION_FILE = wd + project + "a.regression.data_to_train.csv";
String TRAINING_RESULTS_FILE = wd + project + "c.regression.results.csv"; // output of Training(), models performances
String FEATURE_SELECTION_FILE = wd + project + "b.featureSelection.RELIEFF.csv"; // output of Training(), feature selection result
BestModelSelectionAndReport b = new BestModelSelectionAndReport(REGRESSION_FILE, FEATURE_SELECTION_FILE, TRAINING_RESULTS_FILE,
"regression");
}
}
/*
Exit
*/
//System.out.println("##Finished with success !");
//System.exit(0);
}
public static void Main() {
setConfiguration();
}
/**
* set options read from command line
*
* @param args
*/
public static void setOptionsFromCommandLine(String[] args) {
//parse option
String cmd = " ";
for (String s : args) {
cmd += s + " ";
}
String[] options = cmd.split(" -");
//in case of ccmd present (only authorized AT THE END for now)
//this implementation is temporary, need for something more generalized
for (int i = 0; i < options.length; i++) {
if (options[i].startsWith("ccmd") || options[i].startsWith("rcmd")) {
for (int j = i + 1; j < options.length; j++) {
if (!options[j].startsWith("ccmd")
&& !options[j].startsWith("rcmd") //TODO implement: && NOT A RESERVED KEYWORD
) {
options[i] += " -" + options[j];
options[j] = "";
} else {
break;
}
}
}
}
//set options
boolean prefixesDefined = false;
for (String s : options) {
if (s.equals("help")) {
System.out.println("Read readme.md file or https://github.com/mickaelleclercq/BioDiscML");
System.exit(0);
}
if (s.contains("=")) {
setOption(s.split("=")[0], s.split("=")[1]);
}
// get config
if (s.startsWith("config") && configFile.isEmpty()) {
configFile = s.split(" ")[1].trim();
}
// training
if (s.trim().equals("train")) {
training = true;
}
//bestmodel
if (s.startsWith("bestmodel")) {
trainingBestModel = true;
if (s.trim().contains(" ")) {
String modelID[] = s.split(" ");
for (int i = 1; i < modelID.length; i++) {
hmTrainingBestModelList.put(modelID[i], i + "");
}
}
}
//Predict new data
if (s.startsWith("predict")) {
predictNewData = true;
}
}
//if no config file is provided for training, check if we have enough
//information to start biodiscml
if (configFile.isEmpty() && training) {
if (!hmTrainFiles.isEmpty()) {
needConfigFile = false;
}
}
if (!prefixesDefined) {
for (String file : hmTrainFiles.keySet()) {
hmTrainFiles.put(file, "");
}
}
if (predictNewData) {
System.out.println("#### Mode: Prediction");
System.out.println("Merging ID: " + mergingID);
System.out.println("Configuration file: " + configFile);
if (isClassification) {
System.out.println("Prediction type: Classification");
} else {
System.out.println("Prediction type: Regression");
}
} else if (training) {
System.out.println("#### Mode: Training");
} else if (trainingBestModel) {
System.out.println("#### Mode: Best model");
} else {
System.err.println("[error] No mode selected (train, bestmodel or predict). "
+ "Add -train or -bestmodel or -predict to your command line");
}
//export config file when executing command line options
if (configFile.isEmpty()) {
try {
if (debug) {
System.out.println("Export config file in " + wd + project + ".config");
}
PrintWriter pw = new PrintWriter(new FileWriter(wd + project + ".config"));
for (String option : options) {
if (option.contains("=")) {
pw.println(option);
}
}
pw.close();
} catch (Exception e) {
e.printStackTrace();
}
}
}
/**
* read config file
*/
public static void setConfiguration() {
if (!new File(configFile).exists()) {
if (configFile.isEmpty()) {
configFile = "empty";
}
System.err.println("[error] Configuration file not found (provided source: " + configFile + "). Set config file with -config option");
System.exit(0);
}
System.out.println("#### Reading configuration file " + configFile);
String line = null;
try {
BufferedReader br = new BufferedReader(new FileReader(configFile));
while (br.ready()) {
line = br.readLine();
if (!line.startsWith("#") && !line.trim().isEmpty()) {
String option = line.split("=")[0].trim();
String value = line.split("=")[1].trim();
//System.out.println(option + ":" + value);
setOption(option, value);
}
}
} catch (Exception e) {
System.err.println("Parsing error in config file at line " + line);
e.printStackTrace();
System.exit(0);
}
}
private static void setOption(String option, String value) {
switch (option) {
case "config":
configFile = value.trim();
break;
case "debug":
debug = Boolean.valueOf(value.trim());
break;
case "debug2":
debug2 = Boolean.valueOf(value.trim());
break;
case "wd":
wd = value.trim();
if (!wd.endsWith(File.separator)) {
wd = wd + File.separator;
}
break;
case "project":
project = value.trim() + "_";
System.out.println("Project name: " + project);
break;
case "trainFile":
try {
hmTrainFiles.put(wd + value.split(",")[0].trim(), value.split(",")[1].trim()); //filename,prefix
} catch (Exception e) {
hmTrainFiles.put(wd + value.replace(",", "").trim(), ""); //filename
}
break;
case "newDataFile":
try {
hmNewDataFiles.put(wd + value.split(",")[0].trim(), value.split(",")[1].trim()); //filename,prefix
} catch (Exception e) {
hmNewDataFiles.put(wd + value.replace(",", "").trim(), ""); //filename
}
break;
case "validationFile":
try {
hmNewDataFiles.put(wd + value.split(",")[0].trim(), value.split(",")[1].trim()); //filename,prefix
} catch (Exception e) {
hmNewDataFiles.put(wd + value.replace(",", "").trim(), ""); //filename
}
break;
case "excluded":
String excluded[] = value.split(",");
for (String ex : excluded) {
hmExcludedFeatures.put(ex.trim(), "");
}
break;
case "mergingID":
mergingID = value.trim();
if (!trainingBestModel) {
System.out.println("Merging ID: " + mergingID);
}
break;
case "separator":
separator = value.trim();
break;
case "classification":
isClassification = true;
break;
case "regression":
isClassification = false;
break;
case "doClassification":
doClassification = Boolean.valueOf(value.trim());
break;
case "classificationClassName":
classificationClassName = value.trim();
if (doClassification) {
System.out.println("ClassificationClassName: " + classificationClassName);
}
break;
case "classificationFastWay":
classificationFastWay = Boolean.valueOf(value.trim());
break;
case "numberOfBestModels":
numberOfBestModels = Integer.valueOf(value.trim());
break;
case "numberOfBestModelsSortingMetric":
bestModelsSortingMetric = value.trim().toUpperCase();
break;
case "numberOfBestModelsSortingMetricThreshold":
bestModelsSortingMetricThreshold = Double.valueOf(value.trim());
break;
case "ccmd":
switch (value.split(",").length) {
case 1:
classificationFastWayCommands.add(value.trim()
+ ":allopt:allsearch");
break;
case 2:
classificationFastWayCommands.add(value.split(",")[0].trim()
+ ":" + value.split(",")[1].trim().toLowerCase() + ":allsearch");
break;
case 3:
classificationFastWayCommands.add(value.split(",")[0].trim()
+ ":" + value.split(",")[1].trim().toLowerCase()
+ ":" + value.split(",")[2].trim().toLowerCase());
break;
default:
break;
}
break;
case "rcmd":
switch (value.split(",").length) {
case 1:
regressionFastWayCommands.add(value.trim()
+ ":allopt:allsearch");
break;
case 2:
regressionFastWayCommands.add(value.split(",")[0].trim()
+ ":" + value.split(",")[1].trim().toLowerCase() + ":allsearch");
break;
case 3:
regressionFastWayCommands.add(value.split(",")[0].trim()
+ ":" + value.split(",")[1].trim().toLowerCase()
+ ":" + value.split(",")[2].trim().toLowerCase());
break;
default:
break;
}
break;
case "coptimizers":
classificationOptimizers = value.trim().toLowerCase();
break;
case "searchmodes":
searchmodes = value.trim().toLowerCase();
break;
case "doRegression":
doRegression = Boolean.valueOf(value.trim());
break;
case "regressionClassName":
regressionClassName = value.trim();
break;
case "regressionFastWay":
regressionFastWay = Boolean.valueOf(value.trim());
break;
case "roptimizers":
regressionOptimizers = value.trim().toLowerCase();
break;
case "maxNumberOfSelectedFeatures":
maxNumberOfSelectedFeatures = Integer.valueOf(value.trim());
break;
case "maxNumberOfFeaturesInModel":
maxNumberOfFeaturesInModel = Integer.valueOf(value.trim());
break;
case "bootstrapFolds":
bootstrapAndRepeatedHoldoutFolds = Integer.valueOf(value.trim());
break;
case "spearmanCorrelation_lower":
spearmanCorrelation_lower = Double.valueOf(value.trim());
break;
case "spearmanCorrelation_upper":
spearmanCorrelation_upper = Double.valueOf(value.trim());
break;
case "pearsonCorrelation_lower":
pearsonCorrelation_lower = Double.valueOf(value.trim());
break;
case "pearsonCorrelation_upper":
pearsonCorrelation_upper = Double.valueOf(value.trim());
break;
case "maxRankingScoreDifference":
maxRankingScoreDifference = Double.valueOf(value.trim());
break;
case "retreiveCorrelatedGenesByRankingScore":
retreiveCorrelatedGenesByRankingScore = Boolean.valueOf(value.trim());
break;
case "combineModels":
combineModels = Boolean.valueOf(value.trim());
break;
case "retrieveCorrelatedGenes":
retrieveCorrelatedGenes = Boolean.valueOf(value.trim());
break;
case "generateModelWithCorrelatedGenes":
generateModelWithCorrelatedGenes = Boolean.valueOf(value.trim());
break;
case "combinationRule":
combinationRule = value.trim().toUpperCase();
break;
case "sampling":
doSampling = Boolean.valueOf(value.trim());
break;
case "roc_curves":
ROCcurves = Boolean.valueOf(value.trim());
break;
case "loocv":
loocv = Boolean.valueOf(value.trim());
break;
case "samplingFold":
samplingFold = Integer.valueOf(value.trim());
break;
case "cpus":
cpus = value.trim();
break;
case "computeBestModel":
computeBestModel = Boolean.valueOf(value.trim());
break;
case "modelFile":
modelFile = value.trim();
break;
case "printFailedModels":
printFailedModels = Boolean.valueOf(value.trim());
break;
case "resumeTraining":
resumeTraining = Boolean.valueOf(value.trim());
break;
case "upsetr":
UpSetR = Boolean.valueOf(value.trim());
break;
case "repeatedHoldoutTrain":
repeatedHoldout = Boolean.valueOf(value.trim());
break;
case "bootstrap":
bootstrap = Boolean.valueOf(value.trim());
break;
case "restoreRun":
restoreRun = Boolean.valueOf(value.trim());
break;
case "noFeatureSelection":
noFeatureSelection = Boolean.valueOf(value.trim());
break;
case "previousRunPath":
previousRunPath = value.trim();
break;
case "previousRunProjectName":
previousRunProjectName = value.trim();
break;
case "performShortTest":
performShortTest = Boolean.valueOf(value.trim());
break;
case "missingValueToReplace":
missingValueToReplace = value.trim();
break;
}
}
}
Datasets
Models
