{
  "cells": [
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "%matplotlib inline"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "\n# Regularized OT with generic solver\n\n\nIllustrates the use of the generic solver for regularized OT with\nuser-designed regularization term. It uses Conditional gradient as in [6] and\ngeneralized Conditional Gradient as proposed in [5][7].\n\n\n[5] N. Courty; R. Flamary; D. Tuia; A. Rakotomamonjy, Optimal Transport for\nDomain Adaptation, in IEEE Transactions on Pattern Analysis and Machine\nIntelligence , vol.PP, no.99, pp.1-1.\n\n[6] Ferradans, S., Papadakis, N., Peyr\u00e9, G., & Aujol, J. F. (2014).\nRegularized discrete optimal transport. SIAM Journal on Imaging Sciences,\n7(3), 1853-1882.\n\n[7] Rakotomamonjy, A., Flamary, R., & Courty, N. (2015). Generalized\nconditional gradient: analysis of convergence and applications.\narXiv preprint arXiv:1510.06567.\n\n\n\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "import numpy as np\nimport matplotlib.pylab as pl\nimport ot\nimport ot.plot"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "Generate data\n-------------\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "#%% parameters\n\nn = 100  # nb bins\n\n# bin positions\nx = np.arange(n, dtype=np.float64)\n\n# Gaussian distributions\na = ot.datasets.get_1D_gauss(n, m=20, s=5)  # m= mean, s= std\nb = ot.datasets.get_1D_gauss(n, m=60, s=10)\n\n# loss matrix\nM = ot.dist(x.reshape((n, 1)), x.reshape((n, 1)))\nM /= M.max()"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "Solve EMD\n---------\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "#%% EMD\n\nG0 = ot.emd(a, b, M)\n\npl.figure(3, figsize=(5, 5))\not.plot.plot1D_mat(a, b, G0, 'OT matrix G0')"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "Solve EMD with Frobenius norm regularization\n--------------------------------------------\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "#%% Example with Frobenius norm regularization\n\n\ndef f(G):\n    return 0.5 * np.sum(G**2)\n\n\ndef df(G):\n    return G\n\n\nreg = 1e-1\n\nGl2 = ot.optim.cg(a, b, M, reg, f, df, verbose=True)\n\npl.figure(3)\not.plot.plot1D_mat(a, b, Gl2, 'OT matrix Frob. reg')"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "Solve EMD with entropic regularization\n--------------------------------------\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "#%% Example with entropic regularization\n\n\ndef f(G):\n    return np.sum(G * np.log(G))\n\n\ndef df(G):\n    return np.log(G) + 1.\n\n\nreg = 1e-3\n\nGe = ot.optim.cg(a, b, M, reg, f, df, verbose=True)\n\npl.figure(4, figsize=(5, 5))\not.plot.plot1D_mat(a, b, Ge, 'OT matrix Entrop. reg')"
      ],
      "cell_type": "code"
    },
    {
      "metadata": {},
      "source": [
        "Solve EMD with Frobenius norm + entropic regularization\n-------------------------------------------------------\n\n"
      ],
      "cell_type": "markdown"
    },
    {
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "#%% Example with Frobenius norm + entropic regularization with gcg\n\n\ndef f(G):\n    return 0.5 * np.sum(G**2)\n\n\ndef df(G):\n    return G\n\n\nreg1 = 1e-3\nreg2 = 1e-1\n\nGel2 = ot.optim.gcg(a, b, M, reg1, reg2, f, df, verbose=True)\n\npl.figure(5, figsize=(5, 5))\not.plot.plot1D_mat(a, b, Gel2, 'OT entropic + matrix Frob. reg')\npl.show()"
      ],
      "cell_type": "code"
    }
  ],
  "metadata": {
    "language_info": {
      "name": "python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "nbconvert_exporter": "python",
      "version": "3.5.2",
      "pygments_lexer": "ipython3",
      "file_extension": ".py",
      "mimetype": "text/x-python"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3",
      "language": "python"
    }
  },
  "nbformat_minor": 0,
  "nbformat": 4
}