CF_train_predict.java

import j4ml.data.*;
import j4ml.deepnetts.*;
import deepnetts.net.layers.activation.*;

import j4ml.data.*;
import j4ml.deepnetts.*;
import j4ml.ejml.EJMLModel;
import j4ml.ejml.EJMLModel.ModelType;

import java.util.Arrays;
import java.util.ArrayList;
import java.util.List;

import twig.data.GraphErrors;
import twig.data.H1F;
import twig.data.H2F;
import twig.graphics.TGCanvas;

//run with /open CF_train_predict.java
// train_test();

void train_test_cf() {

  int nClass = 9;
  Boolean usingFTOF = true;
  if (usingFTOF) {
    nClass = 11;
  }

  // String
  // trainDir="/Users/tyson/data_repo/trigger_data/sims/claspyth_train/for_pid/";
  String trainDir = "/Users/tyson/data_repo/trigger_data/rgd/018326/for_caos_pid/";
  //String trainDir = "/Users/tyson/data_repo/trigger_data/rgd/018777/for_caos_pid/";

  DataList dlt = DataList.fromCSV(trainDir + "train_cf_wFTOF.csv",
      DataList.range(0, 6), DataList.range(6, 6 + nClass)); //_outbending
  DataList dle = DataList.fromCSV(trainDir + "test_cf_wFTOF.csv",
      DataList.range(0, 6), DataList.range(6, 6 + nClass));
  DataList dlv = DataList.fromCSV(trainDir + "test_cf_wFTOF.csv",
      DataList.range(6 + nClass, 6 + nClass + 3), DataList.range(6, 6 + nClass));
  dlt.shuffle();

  dlt.scan();

  DeepNettsNetwork regression = new DeepNettsNetwork();
  regression.activation(ActivationType.TANH); // or ActivationType.TANH
  regression.outputActivation(ActivationType.LINEAR);
  regression.init(new int[] { 6, 12, 36, 27, 18, nClass });
  regression.train(dlt, 1000);

  regression.save("cf_el_wFTOF.network"); //_outbending

  EJMLModel model = new EJMLModel("cf_el_wFTOF.network", ModelType.TANH_LINEAR); //_outbending

  System.out.println("network structure: " + model.summary());
  System.out.println("\n\nRunning Inference:\n------------");

  String[] dets = new String[] { "PCAL", "ECIN", "ECOUT" };

  for (String det : dets) {

    calcMSE(dle, model, det, nClass);
    //plotCFExamples(dle,model,det,1,nClass);
    plotConvertedLs(dle,det);

    plotCFVarDep(dle,dlv,model,det,0,"P","[GeV]",1,9.0,1.0,nClass);
    plotCFVarDep(dle,dlv,model,det,1,"Theta","[Deg]",5.0,35.0,5.,nClass);
    plotCFVarDep(dle,dlv,model,det,2,"Phi","[Deg]",-180,180,10.,nClass);
  }

  if (usingFTOF) {
    plotFTOF(dle, model, nClass);
    plotFTOFVarDep(dle,dlv,model,0,"P","[GeV]",1,9.0,1.0,nClass);
    plotFTOFVarDep(dle,dlv,model,1,"Theta","[Deg]",5.0,35.0,5.,nClass);
    plotFTOFVarDep(dle,dlv,model,2,"Phi","[Deg]",-180,180,10.,nClass);
  }

}

public void plotConvertedLs(DataList dle, String dets) {

  int nEvents = dle.getList().size();

  int out_start = 0;
  double end = 68.5;
  int n = 72;
  if (dets == "ECIN") {
    out_start = 3;
    end = 36.5;
    n = 40;
  }
  if (dets == "ECOUT") {
    out_start = 6;
    end = 36.5;
    n = 40;
  }

  H1F hStripsU = new H1F("U Views", n, -3.5, end);
  hStripsU.attr().setLineColor(2);
  hStripsU.attr().setLineWidth(3);
  hStripsU.attr().setTitleX(dets + " Strips");

  H1F hStripsV = new H1F("V Views", n, -3.5, end);
  hStripsV.attr().setLineColor(5);
  hStripsV.attr().setLineWidth(3);
  hStripsV.attr().setTitleX(dets + " Strips");

  H1F hStripsW = new H1F("W Views", n, -3.5, end);
  hStripsW.attr().setLineColor(3);
  hStripsW.attr().setLineWidth(3);
  hStripsW.attr().setTitleX(dets + " Strips");

  H1F hLsU = new H1F("U Views", 110, -10, 500);
  hLsU.attr().setLineColor(2);
  hLsU.attr().setLineWidth(3);
  hLsU.attr().setTitleX(dets + " Ls");

  H1F hLsV = new H1F("V Views", 110, -10, 500);
  hLsV.attr().setLineColor(5);
  hLsV.attr().setLineWidth(3);
  hLsV.attr().setTitleX(dets + " Ls");

  H1F hLsW = new H1F("W Views", 110, -10, 500);
  hLsW.attr().setLineColor(3);
  hLsW.attr().setLineWidth(3);
  hLsW.attr().setTitleX(dets + " Ls");

  for (int i = 0; i < nEvents; i += 1) {
    float[] desired = dle.getList().get(i).floatSecond();
    for (int j = 0; j < 9; j++) {
      desired[j] = desired[j] * 500;
    }
    hLsU.fill(desired[out_start + 0]);
    hLsV.fill(desired[out_start + 1]);
    hLsW.fill(desired[out_start + 2]);
    int[] desired_strips = convLtoStrip(desired);
    hStripsU.fill(desired_strips[out_start + 0]);
    hStripsV.fill(desired_strips[out_start + 1]);
    hStripsW.fill(desired_strips[out_start + 2]);
  }

  TGCanvas c = new TGCanvas();
  c.setTitle(dets + " Strips");
  c.draw(hStripsU).draw(hStripsV, "same").draw(hStripsW, "same");
  c.region().showLegend(0.05, 0.95);

  TGCanvas cl = new TGCanvas();
  cl.setTitle(dets + " Ls");
  cl.draw(hLsU).draw(hLsV, "same").draw(hLsW, "same");
  cl.region().showLegend(0.05, 0.95);

}

public void plotCFExamples(DataList dle, EJMLModel model, String dets, int nExamples, int nClass) {

  int out_start = 0;
  int n = 68;
  double end = n * 3 + 0.5;
  if (dets == "ECIN") {
    out_start = 3;
    end = n * 3 + 0.5;
    n = 36;
  }
  if (dets == "ECOUT") {
    out_start = 6;
    end = n * 3 + 0.5;
    n = 36;
  }

  float[] output_l = new float[nClass];

  for (int i = 0; i < nExamples; i += 1) {

    H1F hTrue = new H1F("True", n * 3 + 1, -0.5, end);
    hTrue.attr().setLineColor(2);
    hTrue.attr().setFillColor(2);
    hTrue.attr().setLineWidth(3);
    hTrue.attr().setTitleX(dets + " Cluster Local Position");

    H1F hPred = new H1F("Pred", n * 3 + 1, -0.5, end);
    hPred.attr().setLineColor(5);
    hPred.attr().setLineWidth(3);
    hPred.attr().setTitleX(dets + " Cluster Local Position");

    float[] input = dle.getList().get(i).floatFirst();
    float[] desired_l = dle.getList().get(i).floatSecond();
    model.getOutput(input, output_l);
    for (int j = 0; j < 9; j++) {
      desired_l[j] = desired_l[j] * 500;
      output_l[j] = output_l[j] * 500;
    }
    int[] desired = convLtoStrip(desired_l);
    int[] output = convLtoStrip(output_l);

    hTrue.fill(desired[out_start + 0]);
    hTrue.fill(desired[out_start + 1] + n);
    hTrue.fill(desired[out_start + 2] + n * 2);

    hPred.fill(output[out_start + 0]);
    hPred.fill(output[out_start + 1] + n);
    hPred.fill(output[out_start + 2] + n * 2);
    TGCanvas c = new TGCanvas();
    c.setTitle(dets + " Cluster Position");
    c.draw(hTrue).draw(hPred, "same");
    c.region().showLegend(0.05, 0.95);

  }

}

public void calcMSE(DataList dle, EJMLModel model, String dets, int nClass) {
  int nEvents = dle.getList().size();
  float[] output_l = new float[nClass];

  int out_start = 0;
  double end = 68.5;
  int n = 69;
  if (dets == "ECIN") {
    out_start = 3;
    end = 36.5;
    n = 37;
  }
  if (dets == "ECOUT") {
    out_start = 6;
    end = 36.5;
    n = 37;
  }

  float adifU = 0, adifV = 0, adifW = 0;

  H2F hDifTrueU = new H2F("U View", n, -0.5, end, 11, -5.5, 5.5);
  hDifTrueU.attr().setTitleY("U View Position Difference [strips]");
  hDifTrueU.attr().setTitleX("U View  True Position");

  H2F hDifTrueV = new H2F("V View", n, -0.5, end, 11, -5.5, 5.5);
  hDifTrueV.attr().setTitleY("V View Position Difference [strips]");
  hDifTrueV.attr().setTitleX("V View  True Position");

  H2F hDifTrueW = new H2F("W View", n, -0.5, end, 11, -5.5, 5.5);
  hDifTrueW.attr().setTitleY("W View Position Difference [strips]");
  hDifTrueW.attr().setTitleX("W View  True Position");

  H1F hDifU = new H1F("U View", 11, -5.5, 5.5);
  hDifU.attr().setLineColor(2);
  hDifU.attr().setLineWidth(3);
  hDifU.attr().setTitleX("Position Difference [strips]");

  H1F hDifV = new H1F("V View", 11, -5.5, 5.5);
  hDifV.attr().setLineColor(5);
  hDifV.attr().setLineWidth(3);
  hDifV.attr().setTitleX("Position Difference [strips]");

  H1F hDifW = new H1F("W View", 11, -5.5, 5.5);
  hDifW.attr().setLineColor(3);
  hDifW.attr().setLineWidth(3);
  hDifW.attr().setTitleX("Position Difference [strips]");

  for (int i = 0; i < nEvents; i += 1) {
    float[] input = dle.getList().get(i).floatFirst();
    float[] desired_l = dle.getList().get(i).floatSecond();
    model.getOutput(input, output_l);
    for (int j = 0; j < 9; j++) {
      desired_l[j] = desired_l[j] * 500;
      output_l[j] = output_l[j] * 500;
    }
    int[] desired = convLtoStrip(desired_l);
    int[] output = convLtoStrip(output_l);

    adifU += (desired[out_start + 0] - output[out_start + 0]);
    adifV += (desired[out_start + 1] - output[out_start + 1]);
    adifW += (desired[out_start + 2] - output[out_start + 2]);
    hDifU.fill((desired[out_start + 0] - output[out_start + 0]));
    hDifV.fill((desired[out_start + 1] - output[out_start + 1]));
    hDifW.fill((desired[out_start + 2] - output[out_start + 2]));
    hDifTrueU.fill(desired[out_start + 0], (desired[out_start + 0] - output[out_start + 0]));
    hDifTrueV.fill(desired[out_start + 1], (desired[out_start + 1] - output[out_start + 1]));
    hDifTrueW.fill(desired[out_start + 2], (desired[out_start + 2] - output[out_start + 2]));

    /*
     * if((desired[0]-output[0])>0.2 && i<100){
     * System.out.printf("el nb %d U %f V %f W %f \n",i,desired[0],desired[1],
     * desired[2]);
     * }
     */
  }
  adifU = adifU / nEvents;
  adifV = adifV / nEvents;
  adifW = adifW / nEvents;

  System.out.printf(dets + " Average Difference in U %f V %f W %f \n", adifU, adifV, adifW);

  TGCanvas c = new TGCanvas();
  c.setTitle(dets);
  c.draw(hDifU).draw(hDifV, "same").draw(hDifW, "same");
  c.region().showLegend(0.05, 0.95);

  /*
   * TGCanvas c2U = new TGCanvas();
   * c2U.setTitle(dets+" U View Position");
   * c2U.draw(hDifTrueU);
   * 
   * TGCanvas c2V = new TGCanvas();
   * c2V.setTitle(dets+" V View Position");
   * c2V.draw(hDifTrueV);
   * 
   * TGCanvas c2W = new TGCanvas();
   * c2W.setTitle(dets+" W View Position");
   * c2W.draw(hDifTrueW);
   */

}

public void plotCFVarDep(DataList dle, DataList dlv, EJMLModel model, String dets, int cutVar,
    String varName, String varUnits, double low, double high, double step, int nClass) {

  GraphErrors gU = new GraphErrors();
  gU.attr().setMarkerColor(2);
  gU.attr().setMarkerSize(10);
  gU.attr().setTitle("U View");
  gU.attr().setTitleX(varName + " " + varUnits);
  gU.attr().setTitleY(dets + " Position Difference [strips]");

  GraphErrors gV = new GraphErrors();
  gV.attr().setMarkerColor(5);
  gV.attr().setMarkerSize(10);
  gV.attr().setTitle("V View");
  gV.attr().setTitleX(varName + " " + varUnits);
  gV.attr().setTitleY(dets + " Position Difference [strips]");

  GraphErrors gW = new GraphErrors();
  gW.attr().setMarkerColor(3);
  gW.attr().setMarkerSize(10);
  gW.attr().setTitle("W View");
  gW.attr().setTitleX(varName + " " + varUnits);
  gW.attr().setTitleY(dets + " Position Difference [strips]");

  int nEvents = dle.getList().size();

  DataList pred = new DataList();

  int out_start = 0;
  if (dets == "ECIN") {
    out_start = 3;
  }
  if (dets == "ECOUT") {
    out_start = 6;
  }

  for (int i = 0; i < nEvents; i += 1) {
    float[] output_l = new float[nClass];
    float[] input = dle.getList().get(i).floatFirst();
    float[] desired_l = dle.getList().get(i).floatSecond();
    model.getOutput(input, output_l);
    for (int j = 0; j < 9; j++) {
      desired_l[j] = desired_l[j] * 500;
      output_l[j] = output_l[j] * 500;
    }
    int[] desired = convLtoStrip(desired_l);
    int[] output = convLtoStrip(output_l);

    DataEntry de = new DataEntry(
        new double[] { output[out_start + 0], output[out_start + 1], output[out_start + 2] },
        new double[] { desired[out_start + 0], desired[out_start + 1], desired[out_start + 2] });
    pred.add(de);
  }

  for (double q2 = low; q2 < high; q2 += step) {
    float[] metrics = calcMSE_binned(dlv, pred, cutVar, q2, q2 + step, nClass);
    gU.addPoint(q2 + step / 2, metrics[0], 0, 0);
    gV.addPoint(q2 + step / 2, metrics[1], 0, 0);
    gW.addPoint(q2 + step / 2, metrics[2], 0, 0);
  } // Increment threshold on response

  TGCanvas c = new TGCanvas();
  c.setTitle(dets + " Position Difference");
  c.draw(gU).draw(gW, "same").draw(gV, "same");
  c.region().axisLimitsY(-5, 5);
  c.region().showLegend(0.6, 0.25);

}

public float[] calcMSE_binned(DataList dlv, DataList pred, int cutVar, double low, double high, int nClass) {
  int nPreds = 0;

  int nEvents = dlv.getList().size();

  float adifU = 0, adifV = 0, adifW = 0;

  for (int i = 0; i < nEvents; i += 1) {
    float[] vars = dlv.getList().get(i).floatFirst();
    if (vars[cutVar] > low && vars[cutVar] < high) {
      float[] output = pred.getList().get(i).floatFirst();
      float[] desired = pred.getList().get(i).floatSecond();
      adifU += (desired[0] - output[0]);
      adifV += (desired[1] - output[1]);
      adifW += (desired[2] - output[2]);
      nPreds++;
    }
  }
  adifU = adifU / nPreds;
  adifV = adifV / nPreds;
  adifW = adifW / nPreds;

  return new float[] { adifU, adifV, adifW };

}

int[] convLtoStrip(float[] Ls) {
  int[] strips = new int[] { -2, -2, -2, -2, -2, -2, -2, -2, -2 };
  DataList conv = DataList.fromCSV("LtoStrip_convTable.csv",
      DataList.range(0, 69), DataList.range(0, 1));
  int[] nStrips = new int[] { 68, 63, 63, 37, 37, 37, 37, 37, 37 };

  for (int i = 0; i < 9; i++) {

    // System.out.printf("det %d \n",i);
    float[] conv_det_view = conv.getList().get(i).floatFirst();
    // System.out.println("got conv");
    for (int str = 0; str < (nStrips[i]); str++) {
      // in PCAL V, W take distance from other end
      if (i == 1 || i == 2) {
        if (Ls[i] < conv_det_view[str] && Ls[i] >= conv_det_view[str + 1]) {
          strips[i] = str + 1; // add 1 as the upper limit applies to strip+1 due to inverted order
        }

      } else {
        if (Ls[i] >= conv_det_view[str] && Ls[i] < conv_det_view[str + 1]) {
          strips[i] = str;
        }

      }

      // System.out.println("converted");
    }
  }

  return strips;
}

public void plotFTOF(DataList dle, EJMLModel model, int nClass) {
  int nEvents = dle.getList().size();
  float[] output = new float[nClass];

  float adif = 0, adifP = 0;

  H1F hDif = new H1F("FTOF Component", 11, -5.5, 5.5);
  hDif.attr().setLineColor(2);
  hDif.attr().setLineWidth(3);
  hDif.attr().setTitleX("Position Difference [component]");

  H1F hDifP = new H1F("FTOF Path", 100, -10, 10);
  hDifP.attr().setLineColor(2);
  hDifP.attr().setLineWidth(3);
  hDifP.attr().setTitleX("Path Difference [cm]");

  for (int i = 0; i < nEvents; i += 1) {
    float[] input = dle.getList().get(i).floatFirst();
    float[] desired = dle.getList().get(i).floatSecond();
    model.getOutput(input, output);

    desired[9] = desired[9] * 62;
    output[9] = output[9] * 62;
    desired[10] = desired[10] * 1000;
    output[10] = output[10] * 1000;

    adif += (desired[9] - output[9]);
    hDif.fill((desired[9] - output[9]));
    adifP += (desired[10] - output[10]);
    hDifP.fill((desired[10] - output[10]));

  }
  adif = adif / nEvents;
  adifP = adifP / nEvents;

  System.out.printf("FTOF Average Difference in comp %f path %f \n", adif, adifP);

  TGCanvas c = new TGCanvas();
  c.setTitle("FTOF");
  c.draw(hDif);
  c.region().showLegend(0.05, 0.95);

  TGCanvas cP = new TGCanvas();
  cP.setTitle("FTOF Path");
  cP.draw(hDifP);
  cP.region().showLegend(0.05, 0.95);

}

public void plotFTOFVarDep(DataList dle, DataList dlv, EJMLModel model, int cutVar,
    String varName, String varUnits, double low, double high, double step, int nClass) {

  GraphErrors g = new GraphErrors();
  g.attr().setMarkerColor(2);
  g.attr().setMarkerSize(10);
  g.attr().setTitle("FTOF");
  g.attr().setTitleX(varName + " " + varUnits);
  g.attr().setTitleY("FTOF Position Difference [components]");

  GraphErrors gP = new GraphErrors();
  gP.attr().setMarkerColor(2);
  gP.attr().setMarkerSize(10);
  gP.attr().setTitle("FTOF Path");
  gP.attr().setTitleX(varName + " " + varUnits);
  gP.attr().setTitleY("FTOF Path Difference [cm]");

  int nEvents = dle.getList().size();

  DataList pred = new DataList();

  for (int i = 0; i < nEvents; i += 1) {
    float[] output = new float[nClass];
    float[] input = dle.getList().get(i).floatFirst();
    float[] desired = dle.getList().get(i).floatSecond();
    model.getOutput(input, output);

    desired[9] = desired[9] * 62;
    output[9] = output[9] * 62;
    desired[10] = desired[10] * 1000;
    output[10] = output[10] * 1000;

    DataEntry de = new DataEntry(
        new double[] { output[9], output[10] },
        new double[] { desired[9], desired[10] });
    pred.add(de);
  }

  for (double q2 = low; q2 < high; q2 += step) {
    float[] metrics = calcMSEFTOF_binned(dlv, pred, cutVar, q2, q2 + step, nClass);
    g.addPoint(q2 + step / 2, metrics[0], 0, 0);
    gP.addPoint(q2 + step / 2, metrics[1], 0, 0);
  } // Increment threshold on response

  TGCanvas c = new TGCanvas();
  c.setTitle("FTOF Position Difference");
  c.draw(g);
  c.region().axisLimitsY(-5, 5);
  c.region().showLegend(0.6, 0.25);

  TGCanvas cP = new TGCanvas();
  cP.setTitle("FTOF Path Difference");
  cP.draw(gP);
  cP.region().axisLimitsY(-10, 10);
  cP.region().showLegend(0.6, 0.25);

}

public float[] calcMSEFTOF_binned(DataList dlv, DataList pred,int cutVar, double low, double high,
    int nClass) {

  int nEvents = dlv.getList().size();
  int nPreds=0;
  float adif = 0, adifP = 0;

  for (int i = 0; i < nEvents; i += 1) {
    float[] vars = dlv.getList().get(i).floatFirst();
    float[] output = pred.getList().get(i).floatFirst();
    float[] desired = pred.getList().get(i).floatSecond();

    if (vars[cutVar] > low && vars[cutVar] < high) {
      adif += (desired[0] - output[0]);
      adifP += (desired[1] - output[1]);
      nPreds++;
    }
  }
  adif = adif / nPreds;
  adifP = adifP / nPreds;

  return new float[] { adif, adifP };

}