analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import glob\n",
    "import os\n",
    "\n",
    "import torch\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import yaml\n",
    "\n",
    "from dbo.util import configure_matplotlib\n",
    "\n",
    "configure_matplotlib()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "def plot_average(run_results: torch.Tensor, t_points=None, n_std: int = 1, c='b', ls='-', label=None):\n",
    "    if not torch.is_tensor(run_results):\n",
    "        run_results = torch.as_tensor(run_results)\n",
    "    assert run_results.ndim > 1\n",
    "    n = run_results.shape[1]\n",
    "    iterations = np.arange(n) + 1\n",
    "    if t_points is None:\n",
    "        t_points = iterations\n",
    "    filtered_idx = torch.isfinite(run_results).all(dim=0)\n",
    "    filtered_results = run_results[:, filtered_idx]\n",
    "    filtered_t_points = t_points[filtered_idx]\n",
    "    filtered_iterations = iterations[filtered_idx]\n",
    "    r_avg = filtered_results.cumsum(-1)/filtered_iterations\n",
    "    r_mean = r_avg.mean(dim=0)\n",
    "    r_std = r_avg.std(dim=0)\n",
    "    plt.fill_between(filtered_t_points, r_mean - n_std*r_std, r_mean + n_std*r_std, color=c, alpha=0.3)\n",
    "    plt.plot(filtered_t_points, r_mean, color=c, ls=ls, label=label)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "def load_dbo_results(exp_dir, n_reruns):\n",
    "    gs = []\n",
    "    kl = []\n",
    "    elbo = []\n",
    "    bound = []\n",
    "    for r in range(n_reruns):\n",
    "        gs += [torch.load(os.path.join(exp_dir, f\"run-{r+1:02}\", \"dbo-gs_divergences.pth\"))]\n",
    "        kl += [torch.load(os.path.join(exp_dir, f\"run-{r+1:02}\", \"dbo-divergences.pth\"))]\n",
    "        elbo += [torch.load(os.path.join(exp_dir, f\"run-{r+1:02}\", \"dbo-elbo.pth\"))]\n",
    "        bound += [torch.load(os.path.join(exp_dir, f\"run-{r+1:02}\", \"dbo-bounds.pth\"))]\n",
    "    gs = torch.stack(gs)\n",
    "    kl = torch.stack(kl)\n",
    "    elbo = torch.stack(elbo)\n",
    "    bound = torch.stack(bound)\n",
    "    return gs, kl, elbo, bound"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## RKHS objective"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "run_dirs = glob.glob(\"cr-version/toy-1D/run-*\")\n",
    "kl_divergences = []\n",
    "kl_bounds = []\n",
    "for rd in run_dirs:\n",
    "    kl_divergences += [torch.load(f\"{rd}/dbo-divergences.pth\")]\n",
    "    kl_bounds += [torch.load(f\"{rd}/dbo-bounds.pth\")]\n",
    "kl_divergences = torch.stack(kl_divergences)\n",
    "kl_bounds = torch.stack(kl_bounds)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x180 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "n_it = kl_divergences.shape[1]\n",
    "n_batch = 5\n",
    "t = (torch.arange(n_it) + 1)*n_batch\n",
    "\n",
    "plt.figure(figsize=(3,2.5))\n",
    "plot_average(kl_bounds, t_points=t, c='k', ls='--', label=\"Bounds\")\n",
    "plot_average(kl_divergences, t_points=t, c='b', label=\"KL divergence\")\n",
    "plt.legend()\n",
    "plt.ylim(.075, 10)\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.xlim(n_batch, n_it*n_batch)\n",
    "plt.yscale('log')\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"toy-regret.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Dimensionality effects"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "outputs": [],
   "source": [
    "experiment_dirs = [\"experiments/toy-2021-02-19-105524/\",\n",
    "                   \"experiments/toy-2021-02-19-105618/\",\n",
    "                   \"experiments/toy-2021-02-19-130124\"]\n",
    "n_runs = 5\n",
    "n_exp = len(experiment_dirs)\n",
    "dim_kl_divergences = np.zeros((n_runs, n_exp+1))\n",
    "n_dims = np.zeros(n_exp+1)\n",
    "for i, ed in enumerate(experiment_dirs):\n",
    "    config = yaml.load(open(os.path.join(ed, \"toy-args.yaml\")), Loader=yaml.Loader)\n",
    "    n_dims[i] = config.dimensionality\n",
    "    for r in range(n_runs):\n",
    "        dim_kl_divergences[r, i] = torch.load(os.path.join(ed, f\"run-{r+1:02}/dbo-divergences.pth\"))[-1].cpu().item()\n",
    "\n",
    "dim_kl_divergences[:, -1] = kl_divergences[:n_runs, -1]\n",
    "n_dims[-1] = 1"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x216 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3,3))\n",
    "plt.boxplot(dim_kl_divergences, positions=n_dims)\n",
    "plt.xlabel(\"Dimension\")\n",
    "plt.ylabel(\"Average regret\")\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"toy-dimensionality.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Lumpy 2D"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [],
   "source": [
    "dbo_lumpy_exp_dir = \"experiments/toy-2021-02-21-042935\"\n",
    "config = yaml.load(open(os.path.join(dbo_lumpy_exp_dir, \"toy-args.yaml\")), Loader=yaml.Loader)\n",
    "assert config.objective == 'MixtureObjective'\n",
    "n_runs = config.n_repeats\n",
    "dbo_lumpy_gs, dbo_lumpy_kl, dbo_lumpy_elbo, dbo_lumpy_bound = load_dbo_results(dbo_lumpy_exp_dir, n_runs)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [],
   "source": [
    "n_vbmc_runs = 5\n",
    "vbmc_lumpy_gs = []\n",
    "vbmc_lumpy_kl = []\n",
    "vbmc_lumpy_elbo = []\n",
    "vbmc_lumpy_evals = []\n",
    "vbmc_dir = \"experiments/uai2021_vbmc_experiments_lumpy\"\n",
    "for r in range(n_vbmc_runs):\n",
    "    vbmc_lumpy_gs += [np.loadtxt(f\"{vbmc_dir}/vbmc-lumpy-run-{r+1:02}-gsKL.csv\", delimiter=',')]\n",
    "    vbmc_lumpy_kl += [np.loadtxt(f\"{vbmc_dir}/vbmc-lumpy-run-{r+1:02}-kl.csv\", delimiter=',')]\n",
    "    vbmc_lumpy_elbo += [np.loadtxt(f\"{vbmc_dir}/vbmc-lumpy-run-{r+1:02}-elbo.csv\", delimiter=',')]\n",
    "    vbmc_lumpy_evals += [np.loadtxt(f\"{vbmc_dir}/vbmc-lumpy-run-{r+1:02}-f_count.csv\", delimiter=',')]\n",
    "vbmc_lumpy_gs = np.stack(vbmc_lumpy_gs)\n",
    "vbmc_lumpy_kl = np.stack(vbmc_lumpy_kl)\n",
    "vbmc_lumpy_elbo = np.stack(vbmc_lumpy_elbo)\n",
    "vbmc_lumpy_evals = np.stack(vbmc_lumpy_evals)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x180 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "n_it = config.n_iterations\n",
    "t = np.arange(n_it) + 1\n",
    "start_t = 1\n",
    "n_batch = config.n_samples_per_iteration\n",
    "\n",
    "plt.figure(figsize=(3,2.5))\n",
    "plot_average(vbmc_lumpy_gs[:, start_t:], t_points=vbmc_lumpy_evals[0, start_t:], c='r', label='VBMC', ls='--')\n",
    "plot_average(dbo_lumpy_gs, t_points=t*n_batch, c='b', label='KL-UCB')\n",
    "plt.ylabel(\"gsKL\")\n",
    "plt.legend()\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.xlim(10, 100)\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"lumpy-gsKL.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x180 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "start_t = 0\n",
    "plt.figure(figsize=(3,2.5))\n",
    "plot_average(vbmc_lumpy_kl[:, start_t:], t_points=vbmc_lumpy_evals[0, start_t:], c='r', label='VBMC', ls='--')\n",
    "plot_average(dbo_lumpy_kl[:, start_t:], t_points=t[start_t:]*n_batch, c='b', label='KL-UCB')\n",
    "plt.ylabel(\"KL\")\n",
    "plt.ylim(0, 1.5)\n",
    "plt.legend()\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.xlim(10, 100)\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"lumpy-kl.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Circular likelihood"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "outputs": [],
   "source": [
    "n_runs = 5\n",
    "vbmc_circle_gs = []\n",
    "vbmc_circle_elbo = []\n",
    "vbmc_circle_evals = []\n",
    "vbmc_dir = \"experiments/uai2021_vbmc_experiments_circle\"\n",
    "for r in range(n_runs):\n",
    "    vbmc_circle_gs += [np.loadtxt(f\"{vbmc_dir}/vbmc-circle-run-{r+1:02}-sKL_true.csv\", delimiter=',')]\n",
    "    vbmc_circle_elbo += [np.loadtxt(f\"{vbmc_dir}/vbmc-circle-run-{r+1:02}-elbo.csv\", delimiter=',')]\n",
    "    vbmc_circle_evals += [np.loadtxt(f\"{vbmc_dir}/vbmc-circle-run-{r+1:02}-f_count.csv\", delimiter=',')]\n",
    "vbmc_circle_gs = np.stack(vbmc_circle_gs)\n",
    "vbmc_circle_elbo = np.stack(vbmc_circle_elbo)\n",
    "vbmc_circle_evals = np.stack(vbmc_circle_evals)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "outputs": [],
   "source": [
    "dbo_circle_exp_dir = \"experiments/toy-2021-02-21-000112\"\n",
    "config = yaml.load(open(os.path.join(dbo_circle_exp_dir, \"toy-args.yaml\")), Loader=yaml.Loader)\n",
    "assert config.objective == 'CircularObjective'\n",
    "\n",
    "dbo_circle_gs, dbo_circle_kl, dbo_circle_elbo, dbo_circle_bound = load_dbo_results(dbo_circle_exp_dir, n_runs)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x180 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "n_it = 20\n",
    "t = np.arange(n_it) + 1\n",
    "start_t = 0\n",
    "n_batch = 5\n",
    "\n",
    "plt.figure(figsize=(3,2.5))\n",
    "plot_average(vbmc_circle_gs[:, start_t:], t_points=vbmc_circle_evals[0, start_t:], c='r', label='VBMC', ls='--')\n",
    "plot_average(dbo_circle_gs[:, start_t:], t_points=t[start_t:]*n_batch, c='b', label='KL-UCB')\n",
    "plt.ylabel(\"gsKL\")\n",
    "plt.ylim(0.1, 100)\n",
    "plt.yscale('log')\n",
    "plt.legend()\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.xlim(10, 100)\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"circular-gsKL.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x180 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3,2.5))\n",
    "plot_average(vbmc_circle_elbo[:, start_t:], t_points=vbmc_circle_evals[0, start_t:], c='r', label='VBMC', ls='--')\n",
    "plot_average(dbo_circle_elbo[:, start_t:], t_points=t[start_t:]*n_batch, c='b', label='KL-UCB')\n",
    "plt.ylabel(\"ELBO\")\n",
    "plt.ylim(-5, -2)\n",
    "plt.legend()\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.xlim(10, 100)\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"circular-elbo.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Likelihood-free inference"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "outputs": [],
   "source": [
    "from dbo.util import load_object\n",
    "\n",
    "abc_res = load_object(\"experiments/ABC-result-CartPole-Gaussian-prior.pkl\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "outputs": [],
   "source": [
    "from scipy.stats import gaussian_kde\n",
    "from dbo.util import make_grid\n",
    "configure_matplotlib(small_size=12, medium_size=14, bigger_size=16)\n",
    "kde = gaussian_kde(abc_res.samples_array.T)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "outputs": [],
   "source": [
    "grid = make_grid(-3.,3.,0.03)\n",
    "\n",
    "pdf = kde(grid.T)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 288x216 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "v_min = 0\n",
    "v_max = 1.\n",
    "nx = int(grid.shape[0]**0.5)\n",
    "extent = [-3., 3., -3., 3.]\n",
    "\n",
    "plt.figure(figsize=(4,3))\n",
    "plt.imshow(pdf.reshape(nx, nx).T, origin='lower', interpolation='bilinear',\n",
    "           extent=extent, cmap=plt.cm.jet, vmin=v_min, vmax=v_max)\n",
    "plt.plot(0.1, 0.5, \"w*\", label=r\"True $\\theta$\", ms=15, mec='k', mew=2)\n",
    "plt.legend()\n",
    "plt.xlabel(\"Mass of the pole\")\n",
    "plt.ylabel(\"Length of the pole\")\n",
    "plt.colorbar()\n",
    "plt.axis(extent)\n",
    "plt.title(\"Cart-pole posterior\")\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"cartpole-posterior.png\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "outputs": [],
   "source": [
    "dbo_final_samples = torch.load(\"experiments/lfi-2021-02-21-074832/run-03/lfi-final_samples.pth\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "outputs": [],
   "source": [
    "dbo_kde = gaussian_kde(dbo_final_samples.T)\n",
    "dbo_pdf = dbo_kde(grid.T)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 288x216 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(4,3))\n",
    "plt.imshow(dbo_pdf.reshape(nx, nx).T, origin='lower', interpolation='bilinear',\n",
    "           extent=extent, cmap=plt.cm.jet, vmin=v_min, vmax=v_max)\n",
    "plt.plot(0.1, 0.5, \"w*\", label=r\"True $\\theta$\", ms=15, mec='k', mew=2)\n",
    "plt.legend()\n",
    "plt.axis(extent)\n",
    "plt.xlabel(\"Mass of the pole\")\n",
    "plt.ylabel(\"Length of the pole\")\n",
    "plt.colorbar()\n",
    "plt.axis(extent)\n",
    "plt.title(\"Cart-pole KL-UCB approximation\")\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"cartpole-dbo-approximation.png\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "outputs": [],
   "source": [
    "bolfi_dirs = glob.glob(\"experiments/bolfi-2021-02-21-075*\")\n",
    "\n",
    "bolfi_final_samples = torch.load(os.path.join(bolfi_dirs[0], \"bolfi-final_samples.pth\"))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "outputs": [],
   "source": [
    "bolfi_kde = gaussian_kde(bolfi_final_samples.T)\n",
    "bolfi_pdf = bolfi_kde(grid.T)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 288x216 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(4,3))\n",
    "plt.imshow(bolfi_pdf.reshape(nx, nx).T, origin='lower', interpolation='bilinear',\n",
    "           extent=extent, cmap=plt.cm.jet, vmin=v_min, vmax=v_max)\n",
    "plt.plot(0.1, 0.5, \"w*\", label=r\"True $\\theta$\", ms=15, mec='k', mew=2)\n",
    "plt.legend()\n",
    "plt.axis(extent)\n",
    "plt.xlabel(\"Mass of the pole\")\n",
    "plt.ylabel(\"Length of the pole\")\n",
    "plt.colorbar()\n",
    "plt.title(\"Cart-pole GP-UCB approximation\")\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"cartpole-bolfi-approximation.png\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### LFI Reruns"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "outputs": [],
   "source": [
    "bolfi_dirs = glob.glob(\"experiments/bolfi-2021-02-21-075*\")\n",
    "\n",
    "n_runs = len(bolfi_dirs)\n",
    "n_it = 200\n",
    "bolfi_kl = np.zeros((n_runs, n_it))\n",
    "\n",
    "for i, ed in enumerate(bolfi_dirs):\n",
    "    bolfi_kl[i] = torch.load(os.path.join(ed, \"bolfi-divergences.pth\"))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "outputs": [],
   "source": [
    "lfi_dir = \"experiments/lfi-2021-02-21-074832\"\n",
    "lfi_kl = []\n",
    "config = yaml.load(open(os.path.join(lfi_dir, \"lfi-args.yaml\")), Loader=yaml.Loader)\n",
    "n_runs = config.n_repeats\n",
    "n_it = config.n_iterations\n",
    "n_batch = config.n_samples_per_iteration\n",
    "for r in range(n_runs):\n",
    "    lfi_kl += [torch.load(os.path.join(lfi_dir, f\"run-{r+1:02}\", \"lfi-divergences.pth\"))]\n",
    "lfi_kl = torch.stack(lfi_kl)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 216x144 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3,2))\n",
    "t = np.arange(n_it) + 1\n",
    "plot_average(bolfi_kl, c='g', label='GP-UCB', ls='--')\n",
    "plot_average(lfi_kl, t_points=t*n_batch, c='b', label='KL-UCB')\n",
    "plt.legend()\n",
    "plt.xlim(0,n_it*n_batch)\n",
    "plt.xlabel(\"Evaluations\")\n",
    "plt.ylabel(\"KL\")\n",
    "plt.tight_layout()\n",
    "plt.savefig(\"lfi-regret.png\", dpi=300)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}