{
 "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": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ROOT import TMVA, TFile, TTree, TCut, TString, TH1F, TCanvas, gStyle\n",
    "gStyle.SetOptStat(0)\n",
    "%jsroot on\n",
    "from subprocess import call\n",
    "from os.path import isfile\n",
    "from array import array\n",
    "\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "TMVA.Tools.Instance()\n",
    "TMVA.PyMethodBase.PyInitialize()\n",
    "\n",
    "output = TFile.Open('TMVA.root', 'RECREATE')\n",
    "factory = TMVA.Factory('TMVARegression', output,\n",
    "        '!V:!Silent:Color:!DrawProgressBar:Transformations=D,G:AnalysisType=Regression')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "if not isfile('tmva_reg_example.root'):\n",
    "    call(['curl', '-O', 'http://root.cern.ch/files/tmva_reg_example.root'])\n",
    "\n",
    "data = TFile.Open('tmva_reg_example.root')\n",
    "tree = data.Get('TreeR')\n",
    "\n",
    "dataloader = TMVA.DataLoader('dataset')\n",
    "for branch in tree.GetListOfBranches():\n",
    "    name = branch.GetName()\n",
    "    if name != 'fvalue':\n",
    "        dataloader.AddVariable(name)\n",
    "dataloader.AddTarget('fvalue')\n",
    "\n",
    "dataloader.AddRegressionTree(tree, 1.0)\n",
    "dataloader.PrepareTrainingAndTestTree(TCut(''),\n",
    "        'nTrain_Regression=4000:SplitMode=Random:NormMode=NumEvents:!V')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Dense(64, activation='tanh', input_dim=2))\n",
    "model.add(Dense(1, activation='linear'))\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='mean_squared_error', optimizer=\"sgd\")\n",
    "\n",
    "model.save(\"model.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "factory.BookMethod(dataloader, TMVA.Types.kPyKeras, 'PyKeras',\n",
    "        'H:!V:VarTransform=D,G:FilenameModel=model.h5:NumEpochs=20:BatchSize=32')\n",
    "factory.BookMethod(dataloader, TMVA.Types.kBDT, 'BDTG',\n",
    "        '!H:!V:NTrees=1000:nCuts=20:MaxDepth=3')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "factory.TrainAllMethods()\n",
    "factory.TestAllMethods()\n",
    "factory.EvaluateAllMethods()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "reader = TMVA.Reader(\"Color:!Silent\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "branches = {}\n",
    "for branch in tree.GetListOfBranches():\n",
    "    branchName = branch.GetName()\n",
    "    branches[branchName] = array('f', [-999])\n",
    "    tree.SetBranchAddress(branchName, branches[branchName])\n",
    "    if branchName != 'fvalue':\n",
    "        reader.AddVariable(branchName, branches[branchName])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "reader.BookMVA('PyKeras', TString('dataset/weights/TMVARegression_PyKeras.weights.xml'))\n",
    "reader.BookMVA('BDTG', TString('dataset/weights/TMVARegression_BDTG.weights.xml'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "c = TCanvas()\n",
    "h_keras = TH1F(\"PyKeras\", \"PyKeras\", 30, 0, 400)\n",
    "h_bdtg = TH1F(\"BDTG\", \"BDTG\", 30, 0, 400)\n",
    "h_truth = TH1F(\"Ground truth\", \"Ground truth\", 30, 0, 400)\n",
    "for i in range(tree.GetEntries()):\n",
    "    tree.GetEntry(i)\n",
    "    h_keras.Fill(reader.EvaluateMVA(\"PyKeras\"))\n",
    "    h_bdtg.Fill(reader.EvaluateMVA(\"BDTG\"))\n",
    "    h_truth.Fill(branches[\"fvalue\"][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "h_bdtg.SetTitle(\"BDTG vs PyKeras\")\n",
    "h_bdtg.GetXaxis().SetTitle(\"Target\")\n",
    "h_bdtg.Draw()\n",
    "h_keras.Draw(\"SAME\")\n",
    "h_truth.Draw(\"SAME\")\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": 2
}