TMVA_DNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"http://oproject.org/img/ROOT.png\" height=\"30%\" width=\"30%\">\n",
    "<img src=\"http://oproject.org/img/tmvalogo.png\" height=\"30%\" width=\"30%\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"border-top-width: 4px; border-top-color: #34609b;\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# DNN Example"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Declare Factory"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ROOT import TMVA, TFile, TTree, TCut, TString"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "TMVA.Tools.Instance()\n",
    "\n",
    "inputFile = TFile.Open(\"https://raw.githubusercontent.com/iml-wg/tmvatutorials/master/inputdata.root\")\n",
    "outputFile = TFile.Open(\"TMVAOutputDNN.root\", \"RECREATE\")\n",
    "\n",
    "factory = TMVA.Factory(\"TMVAClassification\", outputFile,\n",
    "        \"!V:!Silent:Color:!DrawProgressBar:AnalysisType=Classification\" )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Declare Variables in DataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "loader = TMVA.DataLoader(\"dataset_dnn\")\n",
    "\n",
    "loader.AddVariable(\"var1\")\n",
    "loader.AddVariable(\"var2\")\n",
    "loader.AddVariable(\"var3\")\n",
    "loader.AddVariable(\"var4\")\n",
    "loader.AddVariable(\"var5 := var1-var3\")\n",
    "loader.AddVariable(\"var6 := var1+var2\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup Dataset(s)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tsignal = inputFile.Get(\"Sig\")\n",
    "tbackground = inputFile.Get(\"Bkg\")\n",
    "\n",
    "loader.AddSignalTree(tsignal)\n",
    "loader.AddBackgroundTree(tbackground) \n",
    "loader.PrepareTrainingAndTestTree(TCut(\"\"),\n",
    "        \"nTrain_Signal=1000:nTrain_Background=1000:SplitMode=Random:NormMode=NumEvents:!V\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Configure Network Layout "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# General layout\n",
    "layoutString = TString(\"Layout=TANH|128,TANH|128,TANH|128,LINEAR\");\n",
    "\n",
    "# Training strategies\n",
    "training0 = TString(\"LearningRate=1e-1,Momentum=0.9,Repetitions=1,\"\n",
    "                        \"ConvergenceSteps=2,BatchSize=256,TestRepetitions=10,\"\n",
    "                        \"WeightDecay=1e-4,Regularization=L2,\"\n",
    "                        \"DropConfig=0.0+0.5+0.5+0.5, Multithreading=True\")\n",
    "training1 = TString(\"LearningRate=1e-2,Momentum=0.9,Repetitions=1,\"\n",
    "                        \"ConvergenceSteps=2,BatchSize=256,TestRepetitions=10,\"\n",
    "                        \"WeightDecay=1e-4,Regularization=L2,\"\n",
    "                        \"DropConfig=0.0+0.0+0.0+0.0, Multithreading=True\")\n",
    "trainingStrategyString = TString(\"TrainingStrategy=\")\n",
    "trainingStrategyString += training0 + TString(\"|\") + training1\n",
    "\n",
    "# General Options\n",
    "dnnOptions = TString(\"!H:!V:ErrorStrategy=CROSSENTROPY:VarTransform=N:\"\n",
    "        \"WeightInitialization=XAVIERUNIFORM\")\n",
    "dnnOptions.Append(\":\")\n",
    "dnnOptions.Append(layoutString)\n",
    "dnnOptions.Append(\":\")\n",
    "dnnOptions.Append(trainingStrategyString)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Booking Methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Standard implementation, no dependencies.\n",
    "stdOptions =  dnnOptions + \":Architecture=CPU\"\n",
    "factory.BookMethod(loader, TMVA.Types.kDNN, \"DNN\", stdOptions)\n",
    "\n",
    "# CPU implementation, using BLAS\n",
    "#cpuOptions = dnnOptions + \":Architecture=CPU\"\n",
    "#factory.BookMethod(loader, TMVA.Types.kDNN, \"DNN CPU\", cpuOptions)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train Methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "factory.TrainAllMethods()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test and Evaluate Methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "factory.TestAllMethods()\n",
    "factory.EvaluateAllMethods()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot ROC Curve\n",
    "We enable JavaScript visualisation for the plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%jsroot on"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "c = factory.GetROCCurve(loader)\n",
    "c.Draw()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}