{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Computing Bayesian Coresets in Parallel using Edward\n", "\n", "*(Better displayed in [nbviewer](https://nbviewer.jupyter.org/) as red warnings in font tag may not be displayed on github)*\n", "\n", "In this notebook we try computing Bayesian coresets in parallel using the approach illustated in Section 4.3 of \\[1\\]. (Notice that by *parallel* we mean logically in parallel, even if the actual computation is sequential). Bayesian coresets are presented in \\[1\\]\\[2\\]. We exploit Tensorflow/Edward \\[4\\] for the definition of models, for their derivation, for posterior approximation, and for sampling. We rely on the code made availabe by Trevor Campbell for computing coresets \\[3\\].\n", "\n", "This notebook builds on the previous notebook, to which we refer for a more detailed description of the code.\n", "\n", "The notebook is divided in three parts:\n", "1. **Setup**: defining the data and the model;\n", "2. **Baselines**: computing and visualizing baselines evaluated on the whole dataset at once;\n", "3. **Parallel Computation of Coresets: Random Subsets**: computing the coresets in parallel when randomly subdividing the data in four subsets.\n", "4. **Parallel Computation of Coresets: Non i.d. Subsets**: computing the coresets in parallel when subdividing the data in two different subsets.\n", "\n", "**ISSUES**: still to correct/review: (i) Rescaling of weights for the model not to crash into NaN; (ii) computing time in Jupyter seems unrelated to the dataset size.\n", "" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 1. Setup\n", "\n", "In this first section we define parameters and auxiliary functions, we generate and show the data, and we define the statistical model." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Importing libraries" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import time\n", "import numpy as np\n", "import scipy.stats as stats\n", "import matplotlib.pyplot as plt\n", "\n", "import sklearn.linear_model as lm\n", "import sklearn.svm as svm\n", "\n", "import tensorflow as tf\n", "import edward as ed\n", "import bcoresets as bc" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Setting a random seed" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "np.random.seed(742)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Defining the projection code\n", "\n", "See previous notebook for an explanation of this code." ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "class _Projection(object):\n", " def __init__(self, N, projection_dim, approx_posterior):\n", " self.dim = approx_posterior.shape[0].value\n", " self.x = np.zeros((N, 0))\n", " self.approx_posterior = approx_posterior\n", " self.update_dimension(projection_dim)\n", " return\n", "\n", " def update_dimension(self, projection_dim):\n", " if projection_dim < self.x.shape[1]:\n", " self.x = self.x[:, :projection_dim]\n", "\n", " if projection_dim > self.x.shape[1]:\n", " old_dim = self.x.shape[1]\n", " w = np.zeros((self.x.shape[0], projection_dim))\n", " w[:, :old_dim] = self.x\n", " w *= np.sqrt(old_dim)\n", " for j in range(projection_dim-old_dim):\n", " w[:, j+old_dim] = self._sample_component()\n", " w /= np.sqrt(projection_dim)\n", " self.x = w\n", " return\n", "\n", " def get(self):\n", " return self.x.copy()" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "class ProjectionF(_Projection):\n", " def __init__(self, data, grad_log_likelihood, projection_dim, approx_posterior):\n", " self.data = data\n", " self.grad_log_likelihood = grad_log_likelihood\n", " \n", " _Projection.__init__(self, data.shape[0], projection_dim, approx_posterior)\n", " \n", " def _sample_component(self):\n", " sample_post = self.approx_posterior.sample().eval()\n", " \n", " sgll = [self.grad_log_likelihood.eval(feed_dict={X:self.data[i].reshape([1,self.data.shape[1]]),\n", " W:np.ones(self.data.shape[0], dtype=np.float64),\n", " theta:sample_post})\n", " for i in range(self.data.shape[0])]\n", " \n", " sgll = np.array(sgll)\n", " \n", " return np.sqrt(self.dim)*sgll[:,np.random.randint(self.dim)]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Defining the params of the simulation\n", "\n", "We define the parameters of the simulation, including the number and dimensionality of the samples we will generate (*nsamples*, *ndims*); the number of random projection dimension and core samples for coreset computation (*nrandomdims*, *ncoresamples*); the number of samples, the step length, the burn-in duration and the amount of thinning for the MC simulation (*n_mcsamples*, *n_mcstepsize*, *n_mcburnin*, *n_mcthinning*,); the number of samples for posterior prediction (*n_ppcsamples*); and a grid step to plot contour plots (*gridstep*)." ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "nsamples = 200\n", "nfeatures = 2\n", "\n", "nrandomdims = 100\n", "ncoresamples = 40\n", "\n", "n_mcsamples = 10000\n", "n_mcstepsize = 0.005\n", "n_mcburnin = int(n_mcsamples/2)\n", "n_mcthinning = 5\n", "\n", "ppc_samples = 100\n", "\n", "gridstep = .1" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Data generation\n", "\n", "We generate data from three Gaussian distributions. We use two-dimensional data for ease of visualization." ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "mu1 = np.array([-2,-2])\n", "sigma1 = np.array([[3,.5],[.5,1]])\n", "mu2 = np.array([2,2])\n", "sigma2 = np.array([[1,0],[0,1]])\n", "mu3 = np.array([-2,2.5])\n", "sigma3 = np.array([[1,0],[0,1]])\n", "\n", "pdf1 = stats.multivariate_normal(mu1,sigma1)\n", "pdf2 = stats.multivariate_normal(mu2,sigma2)\n", "pdf3 = stats.multivariate_normal(mu3,sigma3)\n", "\n", "X_pdf1 = pdf1.rvs(size=nsamples)\n", "y_pdf1 = np.zeros(nsamples).reshape((nsamples,1))\n", "X_pdf2 = pdf2.rvs(size=nsamples)\n", "y_pdf2 = np.ones(nsamples).reshape((nsamples,1))\n", "X_pdf3 = pdf3.rvs(size=nsamples)\n", "y_pdf3 = np.ones(nsamples).reshape((nsamples,1))\n", "\n", "Xtr = np.vstack((X_pdf1,X_pdf2,X_pdf3))\n", "ytr = np.vstack((y_pdf1,y_pdf2,y_pdf3))\n", "\n", "permutation = np.random.permutation(Xtr.shape[0])\n", "Xtr = Xtr[permutation]\n", "ytr = ytr[permutation]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Data visualization\n", "\n", "We plot the data we generated." ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "xmeshgrid, ymeshgrid = np.mgrid[-7:7:gridstep, -6:6:gridstep]\n", "xymeshgrid = np.c_[xmeshgrid.ravel(), ymeshgrid.ravel()]\n", "\n", "ycolors = np.array(['r']*ytr.shape[0])\n", "ycolors[ytr[:,0]==1] = 'b'\n", "\n", "plt.figure()\n", "plt.title('Data')\n", "plt.scatter(Xtr[:,0],Xtr[:,1], c=ycolors)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')\n", "\n", "xcontourgrid, ycontourgrid = np.mgrid[-6:2:gridstep, -6:2:gridstep]\n", "xycontourgrid = np.dstack((xcontourgrid, ycontourgrid))\n", "plt.contour(xcontourgrid, ycontourgrid, pdf1.pdf(xycontourgrid))\n", "\n", "xcontourgrid, ycontourgrid = np.mgrid[0:5:gridstep, 0:5:gridstep]\n", "xycontourgrid = np.dstack((xcontourgrid, ycontourgrid))\n", "plt.contour(xcontourgrid, ycontourgrid, pdf2.pdf(xycontourgrid))\n", "\n", "xcontourgrid, ycontourgrid = np.mgrid[-5:0:gridstep, 0:5:gridstep]\n", "xycontourgrid = np.dstack((xcontourgrid, ycontourgrid))\n", "plt.contour(xcontourgrid, ycontourgrid, pdf3.pdf(xycontourgrid))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Defining the model\n", "\n", "We define a standard Bayesian weighted regression model as:\n", "$$\n", "P(Y \\vert \\theta) \\sim Bern \\left\\{ p = \\sigma(X \\theta)^w \\right\\}\n", "$$\n", "where $\\theta$ is the two-dimensional vector of parameters and $\\sigma(x) = \\frac{1}{1+e^{-x}}$ is the logistic function.\n", "\n", "We also define the log-likelihood of the models and we use Tensorflow for the definition of the gradient of the log-likelihood." ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "X = tf.placeholder(tf.float64,[None,nfeatures])\n", "W = tf.placeholder(tf.float64,[None])\n", "theta = ed.models.Normal(loc=tf.zeros(nfeatures,dtype=tf.float64),scale=tf.ones(nfeatures,dtype=tf.float64))\n", "likelihood = tf.sigmoid(ed.dot(X,theta))\n", "log_likelihood = W*tf.log(likelihood)\n", "y = ed.models.Bernoulli(probs=tf.exp(log_likelihood))\n", "\n", "grad_log_likelihood = tf.gradients(log_likelihood,[theta])[0]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 2. Baselines\n", "\n", "In this second part we compute and plot baselines obtained by running our computations on the dataset at once (that is, not in parallel)." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Performing standard logistic regression\n", "\n", "We run standard logistic regression to learn a model for our data. We run the LR algorithm on the whole data and we visualize the result." ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/sklearn/utils/validation.py:578: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().\n", " y = column_or_1d(y, warn=True)\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 9, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "lr = lm.LogisticRegression()\n", "lr.fit(Xtr,ytr)\n", "probs_LR = lr.predict_proba(xymeshgrid)[:, 1].reshape(xmeshgrid.shape)\n", "\n", "plt.title('Logistic Regression')\n", "plt.contourf(xmeshgrid, ymeshgrid, probs_LR, 25, cmap=\"RdBu\")\n", "plt.scatter(Xtr[:,0],Xtr[:,1], c=ycolors)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Performing Bayesian logistic regression\n", "\n", "Similarly, we run Bayesian logistic regression to learn a model for our data. First, we compute a MAP point estimate for $\\theta$ in order to plot a line discriminator; second, we compute an estimate of the distibution of $\\theta$ via Hamiltonian Monte Carlo in order to computer posterior probabilities. Finally we visualize the results. " ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 1s | Loss: 40.624\n", "10000/10000 [100%] ██████████████████████████████ Elapsed: 10s | Acceptance Rate: 1.000\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 10, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEICAYAAABLdt/UAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvIxREBQAAIABJREFUeJzsvXecHMd17/ut6jCzERuxC2AXORMAE0iKFClRpChKoqIlSzYt2ZLlp2slX9vSux/7+eN0bV/fa/v5yb66sq1oJSpHi0qkLFHMEUQgMoiMxWIXm8P0dHfV+6N6Zif0zM4CELAE+/v57IdEh+rqnpnTVafO+R2htSYhISEh4fJBXuoOJCQkJCRcWBLDnpCQkHCZkRj2hISEhMuMxLAnJCQkXGYkhj0hISHhMiMx7AkJCQmXGYlhT0hISLjMSAx7wgsCIcQRIcS0EGJCCDEshLhXCNF7qfuVQwjxLiHEQxX2/VwIkYn6PiiE+JYQYtHF7mPCi4fEsCe8kHi91roRWAT0A//7EvdnLnww6vtqoBH4h0vcn4TLmMSwJ7zg0FpngG8AG3PbhBB3CSG2CSHGhBDHhRB/UbDvXiHEhwrbEELsEEK8Kfr/9UKI+4QQQ0KIfUKItxUc91ohxG4hxLgQ4qQQ4iPn2fcR4DvAVefTTkJCNRLDnvCCQwhRD7wdeKxg8yTwm0ALcBfwvpzhBj4HvKPg/CuBJcAPhBANwH3APcBC4NeBjwshrogO/zTwX7TWTcAm4D/Ps+/twK8AB8+nnYSEaiSGPeGFxHeEECPAGHAH8Pe5HVrrn2utd2qtldZ6B/Bl4OXR7u8Ca4QQa6J/vxP4qtY6C7wOOKK1/qzWOtBaPwN8E3hrdKwPbBRCNGuth6P958I/CyFGgUGgA/jQLMcnJJwziWFPeCHxJq11C5ACPgg8IIToBhBC3CCE+JkQYiAyoL+LMaBorT3ga8A7hBASMyr/QtTmMuAGIcRI7g/4DaA72v8W4LXAUSHEA0KIG8+x77+ntV4AbAFagZ5zbCchYVYSw57wgkNrHWqtvwWEwM3R5nuA7wG9kQH9V0AUnPY5jMG+HZjSWj8abT8OPKC1bin4a9Ravy+61pNa6zdi3DTfwbwgzqfvO4G/Bv6PEELMdnxCwrmQGPaEFxzC8EbMyHdPtLkJGNJaZ4QQ1wN3F54TGXIF/L/MjNYBvg+sFUK8UwjhRH/XCSE2CCFcIcRvCCEWaK19jAsonKVr6cK/Csd9DvOieMNc7z0hoRYSw57wQuI/hBATGAP7N8Bvaa2fi/a9H/jvQohx4M+IH1l/HtgMfDG3QWs9DrwK+DXgFHAa+F8Ydw8Yf/wRIcQYxr3zDipzEzBd+CeEsEsPinz7/wz8aQ33nJAwZ0RSaCPhxYIQ4jeB92qtb5714ISEFzDJiD3hRUEUIvl+4BOXui8JCb9sEsOecNkjhLgTGMBkq95zibuTkPBLJ3HFJCQkJFxmJCP2hISEhMuMshX7i4Fw6rRMNV+KSyckzEprgwXA8GS1yMaEhIuPmjwzqLXunO24S2LYZaqZ9FV3z35gQsIl4I3XNgLwlacnLnFPEhKKmXr4o0drOS5xxSQkJCRcZiSGPSEhIeEyIzHsCQkJCZcZiWFPSEhIuMxIDHtCQkLCZUZi2BMSEhIuMxLDnpCQkHCZcUni2BPmN+HYMoK+l6H9Rqy2XdhdjyOs7KXuVkJCQo0khj2hCP/0jfiH3wTKBiRqfDnB6ZeSvvIfXxDGXU0sJuh/CTqow2rfidW+EyESPaSEFxeJYU/Io0M3MuruzEblor1WgtMvwVnyi0vXuRqYeSlZgEU4tBl5+gipKz6BEOpSdy8h4aKR+NgT8qjxpSBi9FGUS3h2y8Xv0BzQQbrgpWS0XlAp1PhywsH53feEhAtNYtgT8gh7CnTcV0KjJnqZfvpPyJ64Da3m39cmHF1d4aWUIjh71cXvUELCJWT+/UITLhmi4RTCHcHUfC5Bu+hMB8HxO/H2vvui9202Kvv/FcLKXNS+JCRcahLDnpBHCIw/On0WZAZEFtCAmDlIuajRtajJRZeqm7HI5oMQ50eXPnbXYxe/QwkJl5DEsCcUIdNDpK/5H6Q2fRzZ/DxFRj2PRk0svdhdq4qQitTGfwN7EqxM9GLysXt/gtV85FJ3LyHhopJExSSUIQRYTccJFxxCja8sjpIBQCNSw5ekb9Wwmo5Td92fo0bWocMU1oIDCDfRVE948ZGM2BMq4nQ9Trm/PUS4E8gFBy5Fl2ZFyBCrbTd257bEqCe8aEkMe0JFhDtOatO/INL9IHwQAbL5MKlNH0uSfhIS5jGJKyahKlbTMdLX/E/wm0CECGfqUncpISFhFhLDnjArQgDu+KXuxiVFa0BbCJkUuE6Y/ySGfZ6h/Qb8vptQY6uQdWewFz+IrBu41N160aK1wD9+B8GpWyFMIVJDuCu/jdW2+1J3LSGhIomPfR6hvAVMP/NHBCfuQI2uIzh9I5lnP0w4uupSd+1Fi3/kLoKTt0FYB0i014G377eSzyRhXpMY9nmEf/QuCOpAO9EWG1SK7IFfM66AhDLUVCfZY68me+R1hOPLLmjbOnQITt8CKlVyURf/2Ksv6LVeqKjJRWSP3EX28OsJx+dXbsOLmcQVM48Ih9eTF7AqQGcXgN8IlzB8TytpfMyWf17thGMrCPpvAOVgdWzDansuH2GjpjsITr8E7Tdjte7Bat9R1aft992Ef+SNoCQgCfpuxlr4JO7Kb5p1gfNE+42YzNty1HTn+V/gBU72xG0Ex++M1DQFwemXYnc9hrvyO5e6ay96EsM+jxB2Bh00xe2BOWih6yCNDuoQqZHzDkvUoUv20FsIB68BLRD1/birv4bVdHTObWWPvcq4NZQDSMKhK5At+0mt/yzh2SvIHnhnZKRtwrObESdvJb35Y7EvE51tMmqO+dkNoFKEZ65DdT6D1Xz43G86QrhjFfYoZEPfebf/QkZlWgmOvbrs+Qf9L8Fe+BSy8cSl61xC4oqZT9iLHgTpFW8UPrJ1d01FLnSQIrPnXUw/8d/JbPsjpp/8S4LBzefVJ2/PewgHrwZtAxZ6ajHervehMu1zakd5LQQnXhm5NaKvnUqhRtYSDq8ne/DuKMM1GmuoNHqqm6D/JbHthcMb4rVhlEM4eGHUHIUMsXvuL/9MZICz9IcX5BovVMLhjcTOZpRDcPb8vnMJ509i2OcR9qKHsDq2mWQgaxqkh2w8SWrNV2o639v7LtTwRjOKUi74TWQPvINwvPec+qOmFqLGlxePygCURXDqlrm1NbK2oiEO+l8STedL0C5+/w3xDYqQSm4STUg4sppwdNV5Sww7PffjLP8uwh02L9nGo6Q2/htW07HzaveFjhAKYmeDCkRw0fuTUEziiplHCKFJrfkqaumP0ZOLEakhZMPpms5VmVbU2KoYI2zjn7wNa/3n5twfnemI1zjHRk3Vpu6otSQc2kA4shZ0nONbGENQ2u8c2ebYzVbbc3DobTHNhYT9NxKeyY30Nan1n8VqOVhTf8uaE+AsehRn0aPndP7litW+Aw6/qXyHVNidz178DiUUkRj2eYhMjUBqZE7n6GwLyADCUgMp0VXcJmqqm+zRV6MmliFTQzi9P8Fq3QeAqO+Lap+WXQ3tN6CmO8ti7LWGoO8W/BO3mwXf3MhOWVSaIKqxKtEsOv4rKuwM7tovkN3/TkBFBUKEeXnodNGx3p7foW7rf0+yZi8gwpnEXX2PcaGhMM9e4iz7XpJ3MQ9IDPtlgqxkhEWAtSB+tKomF5HZ8V/zi5kq24K397dxVn0dZ+FTyPQwVvsOwqHNBQqPRp9dT3WTefbDpDf/M7LxVL5N//idBCdfMRMiqK0Sj0mJvjsC/LaK91VNRdJu34W19S8JhzahlY32GwlO3B7roQnPXondnYy6S9Gac44gsju3I+pPm+QtoXAW/wxZP3hB+5dwblwwH7sQwhJCbBNCfP9CtZlQO8LOYC/5WclCXwiWh73457HnZI++Nm/U8ygX//Ab0ZHbxF17j1lAJOeSyVkBC5RD9sgb86dqZRcb9fieVtgWt13h9NxfpS0QzhR21xM4ix4xLcSN8JWFDuqqtvNiIxxdxfS2jzD9yD8y9fhfkz1+R/4zrxX/5K142z9MOHg14cA1ZLZ/xLjcEi45F3LE/l+BPUC8UzThl47V8bRZMJxYCkIjF+zDXfltZCo+bE+NLyP23R4tvOKOoTPtUbWkuDGANIurEbqCP/zcUdid22o+Wrbsh5O3lb9YpMJq2Y/26/GP30E4tAVkFnvRQ9jdj9QUEqqzzYQj60BmsVr31BSlNF9REz14u987MwsLGsxMx2/AWfEd1NhKE3VkZbA7n0Gmy2dNaqIH/9hrzNpIwePz9vw2ddf/2Qv6+VwOXBDDLoToAe4C/gb4wwvRZsLcCMd78XZ9IHLHWKBD1Oha43uP+WGCidOOjZvXAqxp1HQnme1/GPntK4zm7Bm/tXDGKx8303j0Jwv+XekciQ6dmpOiZNMRrNY9xijljLv0TKJT3Rky2/6bSfaKRvX+kdejxpeRWntP1Xb9E68wRkyEpqsaUhs+XXVBVvv16GwLIj0474xc9vgd5W47lSI4fRMq04IaXW+MvggITrwKd/U92J3biw4PzlxXwfWnCYc3YHdsL9+XcNG4UK6YjwL/jdgqyAYhxHuFEE8JIZ7SwfQFumxCDv/wmyNjlgsbtIwcwaG3VDzH6b0vJkY7i7XwCYTlm7T50CEuG9Yc6+EUuHmE5WMv+kWVcDcNMgvumPl/EVDlKxMtvNaupigEuOs+j7vmy8jW55Ctz+Gu/RLumnsIzmw1maSFrhqVIjx7JWq6o2Kb4Xgv/vEoEUelIUyDSuPtfQ+6bKEatLLw9t/N9JN/QWbnB5l+4q/IHn3NvJKE0JVmYEKjRnIvRZEPm80e/HV0WFxFS6sKUUyaCgvuCReT8zbsQojXAWe01k9XO05r/Qmt9Vat9VZhJ/7OC42aiI9V11OLK/pO7Y7tOMvuzcfMI3yszqdwV37btDm2gnijrk090a5HsRf/omiPs+wHyJbdxK5gitBEr2SbMYbDjmLbVczxQSQpUMXwxyCExu7YTnrjp0hv/BR2+06EADW6Ot73L1TFZwe5kWn8MwhH1pdt9Q+/kfDslcYohnWgXIJTL6+YaHWx0Rq0tohPLrJBl5ZBNGSf/xXC4XX575Ldvh1kzExKW1itey9gjxPOhQvxan0p8AYhxGuBNNAshPii1vodF6DthFqxMhA0xmz3qvqQncUPYnc/gvZaEM4Ewp4ZwYvUCDrbGnNWiLPqq4Rnr8Lb+UGs9u3Y3Y8iLD8fiz/91LpyQ6pzi6QF4wntmJG7OwReuwnZ1CAb+nBXfX1Oj6AQNbEYNbUEkR5ENh1GpM+axK/SeHlt7rNyQy6xLzctykamWkmCMzeU14hVKYKTt+F0P3ZuN3MBUePLTRhqmftLQ2oIvDbKxnvKJTyzlXDwSkR6kPTmjyFb9mG17SIc2hQtwGuQIc6y7yGcyYtyLwmVOW/DrrX+Y+CPAYQQtwIfSYz6xcde/ADBiTuKjYrMYnc/NOu5QoaIurNl252e+/H2/WaxgRZZRN0A/vO/mt+uJnsIztxAestHjXF3pkht+BTe3nczY0A0OJOQiRHPkj6ptV9B2BOoqcXI9NmatUa0X4/fdzPh0BXobJNZ9EWa68kA0Ij0EO6aLxH03VIyUA0R7iiyqbKujN2+3UgUlL2kLKyW/cXblBvF08f1M+alewkIhzdWyA1QyJa9qIHrYmY2AuPas9DTXfjHXoO78ju4a7+IGl1LcHYzQmaNRsyLXENnvpA4w+Ypxicrahbxcnp+ivZaCM9cbwyasrHat1XVNNG6evtW226cFd/FP/J6Y7C0hWzZhxpeDxSKP7noTDvhwLXY0ajUajlI3fV/SjhwLcHZLcYnG9SjCSkbASsbmT6LcMeQ9bUnt+hsI9PPfgT8+qg/JQuxkQtFTy/EP34nqY2fJHvg7sjXLpFNR3HXfr5qHLds3VuwIDvLyNTKmAVprzQuXyGbjtR8X79UZGbGJVa0PcBqPIGUAUH/TaAExjyUPBztEAxsxV35HYQAq2V/+Qsu4ZJzQQ271vrnwM8vZJsvNrSy8I+8PtJPcRENJ3BXfWNWbRIhNKnV30Av+wEq02EMZYUpcbb/WoIjbzKuGzmNvfSHOIsfNlN0e7JIKtfuehSVaSU89XIQPmp4A7FRLCpFMHhV3rADqJF1ZJ9/a2RgLeMKKZ3mCx+cCaa3/d8IZwKn56dYnU/VlDTjn7wN/AZmvsYVTtI2angjct3nSV/7V2ivFSF9RA0yyEJo3HWfQ42sIzi7BWFlKo5MhQB35Tfx9v1W5KaRQAgywF3+H7Pf0EXAXvgMwYk7Y13sdscOhP04dvcjBEMbCY6+gdhnGvnZdegQnNlKOLwJ4Yxjdz+M1XT8l3sDCTWRjNjnGd7+d6KGNuQXsfRkL96u95O+6h+QdbNn9QlnCsup/BLw+68hOHg3MwqLdQRH3kxw5HXRJo296CGcZfcihCbov56w72WmP3ljEDfK16iJnpl/aYF38NeKXUPaAQKwMxDURzMLCdkFgEQHjWQPvRU704q79L5Z7zUcuoKav8Ja5GuWigrhn5UQAqzWfXmphWpYbbtJbfo4/vE70NMdyKZjOD33VZ2JaC3R051gZZCp0Tn1ba7I1MiMFEAu4kgLUuv/HWGbaDVZP4Bb/4DJ6B1fSdHLWARYHc+iQ4fMjt83chUqBSjCwatxVn4Dp+vJX+o9JMxOYtjnEcprQQ1trCDk9QpSq2tbTAxH1uCfuB2dXYBccACn56d5g+Ef/hXKg6EE4OYjD4O+m02K+NIfGs2XWJ9rKQLCOtRUJ7J+AJ1pgzAuA9VGWB7p6/4C7+BbjU+3KPM1RXDylThLHpg1/ls4E+jMwqrHRI0iG09etHhyq+ko1sZP1XRscHYT2YO/ZmY12kI2njBGtoIWvPKaUSPrQfpYrbuLFrtrxe7cjtW6l3B0DQiNtWB/Ua6ACi2ye98TGfUocB8B0ria3GX3EvTfaETi8i9uabKWn38Ldsez512QJeH8SAz7PEJPd1QQ8rJQk4tnP9+vJ3Pg7ejhTeSMZTjdQTh4jRnxp0YgrJ+9IypFcOI2ghO3Uf4SqIZAZRcg6wcQlldxIRE7g5AherxCOKUI0dMdiAINmjis7odR4yuIf9FExkhkQYZYnU/i7XsnoLEXPols2XdBqiydD2pykRExK5jVqPGlZJ77XdJX/V1Z//yTt+IffW0UImruL7X+szXNJEoRtofdvit2X2bbH5kIpaLnqrF7f4Sz+EGEVISDW8qjfyAKH12KteDQnPuUcOFIDPs8QtSdqZDcEWA1Vvex62wj09s+AkEUI57HhgD8468ktfob0cJZLR+7LGinWnZoyVl1RmZYuBPI5sOo0ZUUfc2kh7P4AXNM3SA600l5eJ2NqMEloaskFgGIhkNYbXtQ04vwj75uxmUwtAl74RO4q75V0z39svD7bo6JkbfQ020Ep25FNpxELjiEEAo1saRCCv+7sTq3ocaXIdxR0AI1sRykj73wCZxlP0RIkzCWE0oTznhsyUEdpPD7bowx6gY1tAXRYz67wozj4kYs/P5ryB5/JWQ6AInVtgOn9/5zCoPUGtTwRvzTN4JysTufMWswVUomJiSGfV4hU2NYHc+aBJf8aEiBDLCX/Lzquf6JV0LQQLwBtlGjRpxJdjwTuT9mM9QlCox5454bLUpKR3QiNYxMzSxIptZ+geld7zc/cKFASeyux7A6jQ/W6fkp3uia4pGfyJo6qDUYgXDw2gr3oaHhKOkrP4aaWGrCQHXOLSTNjOTMDdjdj9Ssd1/UerR4WC2iSGtBOLSJcOAqsLLYXY9jNR8pPsZrJT5G3o1G5iZTN73pXwjObI1/6WuX8Mx1mBdCd7RRmHvsuxk1uYTUFf+Gf/S1JtwTYdxsPT/BXvKz/KzAP32jKTWoKn0vBGp8OcprQaZGcBY/hDe6tsRNp0A5qIGXUCjsFpy+mfDslaSv/l/oME1w8hWEI2sRqRGcJf9ZVZrBP/IGgtM35a+THV+GPHMdqU0fN8U+EmJJDPtFRGVayR5+k/GRihC780mc5fcW+X7d1V/BT50lOH0zhGlk02Ej5FVlwU95LQQD11Lt48z5bFOrv8b02KooEaWQGvRb7HFk01HsRQ/hH3qbEf3SNvnIjw2fNGdrYeLLj7xhxqhriUgP4fT8FJSDliFW82HcNV8k+/xbzGIqYHVuw135zagdiRpdjQ5TWM2HyvXUK8kNCEVq/edQY8vwdr8vPptSS8LhDXMy7NqvJ/v8WwnPbjbhks2HcFd/rWxRW2uBt+c9BdmuinDwGkR6wMxQhMLqeBrZfMgURylzaYiZkblyyOz4ENT1UfllbM2cV9QRFzW+guyhXyEsiU/3j78a7Cmc7seNoNfhN8W7Vkr65R+5i9S6L2G17MfuuZ/g+KuifZK8Jn5ZP2x00IB/6uUEfS8zay/aRk8vwhtbVXHBVWVazXpPSV1VNdFDOHQFdvvOWfr74iUx7BcJHaTJbP+DaFRtDGjQf6MZUW3+WH7kJKTCXfZj3GU/rqldv/86/OffWl2fQ2ZxlvwnAOHQRsg2Ufzj86NIlVxGYowBsTzc1V/L+2Wta/6WoO8m/JOvMElBWpDd8zvIjmdM9SK/nvyoPhrY6kwn00/+udkuFFbHM9g99yHq+tBja8wxXgs6aEBP15N57nejH7UGZeMs+z7OkhkJA7v7MTOyLRk1irrTCJUms+v3Kt+PCM0914jWkNn1AfT0wrwrS42tIrPj96m79q8RBW2p4Q0lEgZmYVFPLc73JTxzPaKuH+yJ6PnljFeMXr1KweTSCj2bZealJGH/S8uPUy7+obfhH3utiW2vSd9FRLVOo385k6a/2qqhHy7BmeshSFFkdpSLf/jN2J3PlLlX1OjqaFBQ2laKcGhjYtirkBj2i4TRHHEpTadXkz2oid5ziv/V2Ub8Q79aHkUzcwSgcZZ9H6ttNwD+kTfGjGAdCKr8OEWAcMbybZhtoZlV+E0YNUkL7bUTnnxl5XaK7t0iHLjWZHVqSW7UqUZXk3n290FIKFGe9I/ehWw6itV8FAC7+yHCkbXGAGgJMgSZJbX+3/EOvo2KRt3cgNE7KUFNd6CmmyGoI7P9dpze+7HadqPGVplIn6L1CQnKIThzHc7iB/Nbg7Ob43VpCvuiHXSmw2Rvji8nPLvZKHFWPK/QZaNL9lWjmtGV4DcDpS/6KijX1JEN6/Cff3OV714pAfh1xJocLdCZDkR9f/F2e4rYOrkECGf2HIQXM4lhv0ioid4KP1qNnloE52DYw+GN0WJo6Y9LY0auZ0hd8W/5UEetRUxWZI5KP2yN1fEM7orvFvk01dhKk8Jf5iOeS6iJZEY/JodlxLPiImq0Q/bky6hr/oK5klSkN36KcLwXNWEWD63W3SbiZqK3Sl9CUus/W+baUdMdZLZ/GLZ8GRCoiRV4+34TZ/n3In96XGKWG+nVzyCsDGYtYpaIIm2hs624y++F5ffiHXwbYX/OP12NWp9xrYvehWsos5yvMS9jQYXvXgVkaBbL42Ye2kLY5WsqVuveeMMuFXbX47Vd90XKBauglFAd0XDKSNbG7as7U/VcrQVqchFqamGJ/GulxTuN1fkUddf8XVHCixAaYn5AZmechK5CNh8itfbLZUbQLPz9ktCVpIIFeugq/JMvK9pqNR3HWfSQUXLMTefdSlE1GmvJT2MVCP1jd5Zrz6uUiahJ95cdD4D0kI3FL2W764kK1y69nRCZnklcspfcR+XPtBo5jfuyC8yhjdzCeK69sEKbNtnDbyY4XekFpIHcd8lIMwt3hNSGz+As/Um5TLTwkS37Y7OAhQxJX/FxE+0jM5EKaQZ3zZdqStZ7MZOM2C8SzsInCI7fgXnk0ftU+Ii6gao6IuHIarz9v5k3OMIdI7XhM8j608Y1cuhXy0+SAc7iePEvp+c+41ctnD1ID3vJ/QSnboukWx0QPsgQd2V8SKBoPBb5Vmul9pBJyAKVFvIk/rG7kI0nsBY8X7EFp/c+/AO/RvFX3Bgad8kDseeo8ZXER6nISMemDzW5pGCUGpoqQwuLFatlQx+IDOgqOQMiQKRGTdWnCKtuGKv3R4QnXlXgt67luQlilSvnRKR9r6NFdKEqfL4agkbUWElGar4rPvaS/0SkxpCNRxF2BpEayq8h6aU/jIqWKKMRtOAgqbVfnGldC/R0pznPHUM2niK99S9RE0vN8U1H8+GbCZVJDPtFQjhTpLd8lOyht5kfhVBY7dtxV31z5kufbQYR5kP9lNeMt+d3ioywzrST2fkB6q77C4QzibP6q/iH3l70g7QX/7yiOqK9+BegJf6JV5mXhQyx2p/FXvwAdtcTUYhcD7LhJPaih2JT3P0zW00UhY5UFPOGJ6C6T7cUbUZw2p7xXYsA3HHItlSOt1c2Qd/NVQ273fk0wcAW9Mjmgv4YQ6mmumPPFe5IvKtKS6Q7QWrTv+AfeR3BwFbQFrJ1N+6K78RmtIq6QfRUpQVPjWzbRWrVN8pCJlNL7yNs3U/Qfz0oB1F3huDYnVR9rjKLbHrefK+K1k9KXwqF1avi3Cwzax35RevyO4v+mzP6GoQHQoCW2L334fZWrlPrLHnAyERPd4EzXvT9Mlm4bzdia5FIW2rd5xDuBFbT0YptJpSTGPaLiKwfIL35/0Rx0Dpv0NXEErz97zSLcwhkwwncdV8gGLg6xtcsQduEwyYqwFn4NNaCg1Hsu4XVtquqLokQ4PT8HB3WmcLTQDh0JeHQFlIbP4m7/Adm2/hS/COvR2dbkK27cbofQdgZwpG1+IfeWh6/LEJE8yH06LpKV47+O2MsZNuzuKu/gX/0tYSDVwMCq2Mb7vLvk9n5QfRUN5Vqrc4mg6smekqMeu7yLv7R12Nt+aeyc4xM8btKLpXFatuZj3pxV30Lq3Uvft/NaK+FYOAanEUPlxl3Z8kDZA+8o/z6gGg4Tnr95yr23Wo6WmTI7PZnyez4Qwhz2vAFRllkEe4ozuov4T37xzFlDIMoLFQiF+xbWF6vAAAgAElEQVQzn+HkEhNy6DebkbM1HWm0x62XaLC86Noxn4XM4vT+CGF7yNa9Jrt5FoTlI0oGHmpycXkW7thyMrvfS91V/zhrmwnFJIb9ElA4StN+PZldHzALhhFqohdv54cQrbvip9dKRguXBpkaRZZUMqpGOLqK4NTLKcti3P071F3/ZwSDV5tom0ihUI33Ep5+Kemr/sH4ocsWgc0iqC4obF0ZATKLu/bz2O3PAZiM2NXfMLc23YGa6sZZ/h2ye38nksotMY4iC6kRMnvejbAncRY9UjRDycWRVxrhqqnu2O1W2x6cFd8m72sWPlbbDtzVX8sfkz12p3khRs8gmFxMeOZ60lf+Y5E+ir3wGbLHXlOexSm9im6yoj5OdqMmlhq9+Jb91F37N/inXmYSe4SZGWktsdp2Ync/hDp7ZfR5lYRKyhB39VeRC/ahJlagvQU4S36GbDyFDlLoMA3OGJlH/75ST3CWfRf/5O3gxWT6aond9dR5F9fwT90Sk4Vro6cXoiYXJTrvcyQx7JeY4MzWGF+mhQ7TWPY0SmZApYt3C7CaK7shZr1m/w2RwSxFEA6vxX/+LSWqjC7aW8D0zvfBVE/MeVGnSvtZCeWYKKHIsANov8Ek9UwuMSPMME3FSA3toAa2ACkgNDrwy76Pu8SEHOrJRRBU7otwKyd7Od2PIZsbQdnUXf9nRfHpOttIcOL24petdtFeK8GZrTiLHi1qK73pX8xL24987drC6tiG1flU5UejJN7ed6FH1oHQIDTCniS1+WO4y34Iy4r19dV0O96OPzQupDifuHJQ40vJHn6jeRkpiS9Atuwjtf7fkbZnZpDpQcgspPx5C/wjb8Zqf4bwbFPxS10EyOZDF6RiknGBxekGKVOAPDHscyIx7JcY7bXHZ/wpC+GOI+v7UVOLZo6RHlbbznNKhc9fUzlUCogyYZlxi2ZWZNRr8Z/PtuAXmtjtsWX5mHRv72+hxnsp/kpWUJFEY4x61C9tERx5M+rslbjrvmBqekZS6OUonGWVi4/kkUGRUQcIx5fHi7SpFOHwFUWGXU0uIhi8EqvzCWR6GCECZNNRZEylqhzB4BayB+6OPuuCxK7Qwdv7W6Q3fhLsqZk1GQ3ec++LIpQqBbjpvLZ//hgNamQtQd9LcRY/aHTts63EP2+TXBWevQZr4aOE/TdF5QstZP0pUuu+GHPO3JEt+1Fjy8tzLJSNbKitmlbCDIlhv8TIpsPQf33MqFyZ9P2uxwhO30QwsBUhAuzuR7AWVh7xVUJnm/DPXAt+MzI9gJJeuUsldAlOvJLKX4u5hM/ljHvO11PoY7dQo2vxnluO3fUY9pKfmVqcNX8d4w2+Gl+Ot/NDpK75H1H4ZnmfZMdT556xaE+CijOgYZHMrnHX3JZ/QYYyxO79MXbdMxWb9g6/lvBUpeQuCz2xjOkn/xLhjuKu/hpWy37U+DK0P5PJHI8V9bk08zRFcPom7EUPGhXP2eQERIjduh936X2oySXGRVSaUIRJzvKP32Eko5uO4C77IbJ+9kGI0/0IYd/NaF/OLJpLD7v74ZoKoiQUk8SxX2Ks9p2mmLIo0K+WWWTTEWTjMYTlY7XvxGo+CLaH8tqiqkG1E46sZvrpPyE49hqCU68gOHWLGXXlY4oDZuKWY3zaVakUdy1K/lto4KM/lTKyCmMrTNboeV/TQmcb0KNrSa37nLk/kcUs7mYQTQdJrflahXNnuaJfj384l2lZcn0Z4kS1ZdVUd2TUcwudlkmnP/ZqVKY9tm011Ul46naqP3cB2kZ77ZHLalEkT1FL3HuFdrUFyKL1naptWBmEM4nVsj/WqPunbia7/zfQk73gN6OGNpHZ/l9RU7Nr5gt7mvRV/4C96EFEegDReAx39ddw5knlqRcayYj9EiNkSHrLP+EffyXh4DUgQqyux3CW/BwhTHSKt+v90ejPRo2uJDh1C+6aL2M1Hy4XxipBaxGVaisYnes0BBrcYWTTbpAeauBaiuqY1n4H57ENk5I/uop4AzWX2PfcKTYq04az6BGz4HjmWvCbsFoORBrs55IABN6ht5kM4aKxkEnGcVZ9DRlpx4dnN1VwZZl9cskDBcqPV4P00WHkeqkVZeOfuA135bdrEO6qhEYrC0IHkTqLjlsYnbmgGWw0Vy76rZVVnh8RuXGyx15TNQooh3AmcVd8D1Z8r/bbSIglMezzAGFncFd8H1Z8v2yfiestNMoOhDbZve82sfALH8dd9a1Yg6UjX2q8wJOAbCvqbGs0qj2Xr8JcUtYro85uwVn5rSgSJzdjyN2PCdUri/aocq1cJqhwx3F7fl5D/6qjlYUauiImrl6AM41TmKAklFn0LPs4tFkI1AJv77ujz8UoP879BSYJhzbh152JkopqTRQrvI4Av4ns87+Cs+I7ZPf/ZslLQpvvhTSy0ekr/rW6TLHXEhOaa/qqx5fV2L+EC0Vi2OcpRtelBT3VFbM3EoXSFuHAdfjuOO7SnxQdEY6sIXvw19BeE7P6zHWchs3FQkCQxmo+hOp4mvDM9RQn48T4hyuiwR2JFVTTWhh9m0w7suFEfoRdW7NVkoNKoous9u1GErfMBgqs9h2okbUFRh1mlUquhEoRHH/13M4pi+l3CAevxl3zVVIbPol/7LWo6U6EM4FsPois70fWn0EuODir9rlwJioYdhCpudWYjUNNLiZ77E70RC8iPYiz9MdJlaYqJIZ9HhIMbi7QKJ9lGUSlCPpeVmTY1VSnieOOFR2bjVoNjC74K3VPzNF9IjQ6TBlXVKUkmRr7lNr4ifKt2UYyuz6I9haQ8+/L5oOkNnxm1ko8OqhD+w1GWrgs1FNHOQXN+cVTWXfWxH0feaPZL8zzcVZ+HZkaxTt+RwX3Seki82xUU66cQ9vaIjhzFQRNyI6nUIdfj1YthGeuJxRgdz2CXLC//LzS3tgeVufT5jMsvD/p4fTOXpi8GuF4L96uD0QvUYnOtuLtXoa75ovYHYl0bxyJYZ9nhGMrTMZi6bS42o84KF788vteXsH9UouBzKkS1qIwKICgIDPRRLxUvk78duFMgzUFutI1azPq9vLvYTWUC6p5B+5GT7dT+HVXY6vxT9xeNtMpvGZm77tMcXGhMM9FUWxQzYKm9/ybi3zIzuKHsdp3EQ5tMv1q34VwxwjHlxAOXBt/OZmNQv5WRp/nuZUmjG97yixOB3HyvAL/4N0F9yWKtMCC/huxWvfG1lVVmdYo49nBan0Od9U3yApNeGareaFJH2f5d2MF1+aCf+T15YOUSMfdat95yWvXzkcSwz7P8I/fEWOUc6PW0tGxoTTOV093Eq+OWAtGayY4dSu1GXgbrFHspfcSHH19hRcKzPS/POHIWf0lpDtRwTddeL5vfL7KMcY2lygkfMSC/flaqkVnhS5qdA1lX3XlEvS/pKJhV1NdxqgXZefG9d9Cnd1CZucHkQv2G3kBZ9JkAy96OGprIdM7PliQuRk/cra7HiNwxlBlhTHO1XJF6xOqriBMM3cPubULcw8VUQ5B//Vlht0/szWqBWA0YvwTr8TueoTU6q+hV3wHHdQj3LELUr5OTfbGbtd+k5FGsL3Y/S9mEsM+z9BRAeAyZBar6xFTDUflFCLNwpZTosBoSq4tL0/2mA3pYS/5Ge7SH2O17sHb/d7a2tAO4elbqrh+cguEcZmFPiAQUiEXPoE6fTOVDJls24Pd9SSy+SBh/0tMRR6hsLsexe5+LHbkppVdeSYQ+ceVtwD8JkRdv5EF0JZxg5XJOVSeUaixVajxXoLTN1N31T/kXTPKW2AqZ6lUlfNNFaLsvt+Kl1A4J4zyJGGpO2+uaxfSJHxh9HfUdCciNRwtdBd8N5RlRvcd27Gaj8QKo50rwhk30gdlO0IokHFImCEx7PMM2XSEMBOXXi1wen+CmuxFj62AqJgG0keUVIx3Fj1McOplEFoz7UgPnAnwWkrajkLZGk8ahccoecdqOYSz6hv4z/9KdFglwxRite00U/JYFFbPj1Hjy9GjG8p3a4fs3nfju2NVioAYrAUHsNuMDIFc8gBOBfndQvzDb6rQ7wDZupvMrt817g8ZgJY4y74PV++m8tQhbtSeW4R2wZdkj99BapWp2xr03TKTSVqRmbj+2uLSayAvy3CeL4ko0zmz4/dQk4vNtkovH2UWY0uLdp8vds99kcxFodR01iQvJQWtY0kSlOYZTu9PolFIwRc20ksPB66LKgPlojRsCOqKFAmVt4DMrveb0aaI3B/2GHbP/Tg9P4nKjWXy7eKMk77mf5He/LGyjEyn60nqbvhTUps+jlywrziJCkzbzjju0h8hFxws7nMOexJ36U9wup6IwipLEaDq0JmuaIRcwRCJAKttT+UHF4PKtJpqP2Vfc22KnnitkdStY5J0oqIauqI7KTAFH6oaXxtVUBdUjS+NuX41ZjPEtRj+AFF/Ovr8z+c6AbJlL2pkPWq8J9KaSTG3SKXzx174JHbPfeb7Kj1j1Bc+jrP83ovWhxca5z1iF0L0Ap8HujG/7E9orcs1URNqQtYNGt32o3cZo2OPY7U9hw7qTcX2GGVFPd2J8loQ7gje7vdGIZIFo/KgnuD4nSB9E5LmTmK1PI3VdAxR10c4cDWhncFu316W8CRkgNV0HLHhM/jH7yA8/VLj13QmsLsewen5OcLK4i67l8zIukhHxSZvgIJGph/7W9z1n0Kkh4wO95wzWzWy/VlkemhOz1JN9sRruyCQjSeNjEGpu0Wl0F4LIj1kjEj+eYdgZXHWfh5/73ti3DQF2NMzV6o7DWMVCnhUZLbF0mqL0wrReJz0xk/j7XtntL4w1/Fb9Nk5Y4i604Qn76itDRlgdT49+3FzRAhwe3+Ks+QBtLcA4Y5fUFfP5ciFcMUEwIe11s8IIZqAp4UQ92mtd892YkI8sr6f9IbPoLXE2/PbhKdvru57FRqUhZ7qRmfaKTcilvEzh9HH7Vlor41QQPj8W4yxFyH+4TeRWv+ZooWyYOBK/CNvNIWW7Smc3h9jL/5FmT9b1g2SvvrvyOz4PVMko7CvKk129+8iGo9i3v1zMXLGTaGGtuCfPIWz5Oc1nylTwxViqwMj4zAZxhh9QFtGk2XtlwhO3I72m5DNB3GW/hjCFH61Wp/Sw1ls+qhDFzWygdgZQ6yODlQfkWtwh6oIdpltemoxKtOGu+bLZHZ8KPo84vzslYj2+22zFCcv6bOyCM9uNlIYv4QBvZABooqIWsIM5+2K0Vr3aa2fif5/HNgDLDnfdhMgHLgKNbp61umvcMYR6bPooIH4qu6l59mo0bWE/VuN/1fb0TTbxdv3rrwrIhzaSPbg3eicIQka8I+9Fj8q0FF+mSxkc7HipVjoibmOXAtQKfxjr0EHtcfmi4YTpp5sqSCYDM3UPpYAEY247fadpK/8KHVb/4rU2i8j00OI+tNlqo8GUyfU7n4kL8vr991kqmLFRAKJtm1YXQ9HfSs05lXi02UWd8V3zcyrIpZ5VoffjLAypi/SI+damTu1VMTK9dkiOHUrfv91KG8B2SN3kdn9HrLH7ojEyhIuFhd08VQIsRy4GkhKiF8AgoGtFSJNohGfNAJX7rrPI0QU9lhrerkWzEjfFm/P7Hof+E3obGN5Mo1KEZx4ZaRlUzy69A7k4qHjKDdulfdVaiJETfbUnHEoBKSv+Fe8A3ejRtYBGpEawV39Faz6AZzl38PP6ZSDMbJWpmqmpBAad/2/4z33u+YZKsdI/Nb1k9rwSWRqPH9sePbK+Kgiy8Nd/BCy6RjTZ66r7f5FgGg6TPboXWXZrnGo8WVMP/43FA8IciXxfolom+Do6wmOOEYzRzuRRPDLSV/5/yHTyYj7YnDBDLsQohH4JvD7WuuxmP3vBd4LIFJNpbsT4qhUtFf6WK3PIZuOYC98Ol/oQNgezrLv4x+9q4ZIjAooN6qEVMUIKDeKH54ZuarpTvTYmjlcUzCTDFUj2kI447MfV3gVZ4r0xk+Zkb5ywJnIuwmc7keR6bP4J1+B9lqwWvbi9PyswqxnBqvpKHVb/5Jw8Gp0tgm54JApOFFy68KZjH+CWiLsaVPeT4gabK1ZZ9Cjq5n5yc6WqSqIf7bn4iOZW8IZQWPxdu1CYJM9/AbSGz57DtdPmCsXxLALIRyMUf+S1jq2rL3W+hPAJwCsxq5f8rDh8sDuepxska5IhMzirvtiPtRLZVpBpRB1/TiLH0TW95E9/KYYNcIcuWShOEoyK+OwPPNXgJpcbELsYqlQCWlOo8cQUWe0S84FYXtAeSKL1bIfq8WkzGslCfpuQU0MgBZkD6/C6b0v75oJx3sJB64BBFbnNuzumcIaWguCs1cQDlyNDuuQzQexFz5GdmQ1xVo8CpEaRtT3RbOrWsL1colbpUVIzifq5WIjUSPrazpSa6OtT1CHbDoyq4JpQjkXIipGAJ8G9mitk6qzFxCr9TnshU8QnLkhso1GOTC14VMIoVBeC96edxsDLkKQAak192C17cFd9U285/5LjCsnJOcPPSekh7P0B2VuGJkeqLBQCZWNzGyj9SgsEYGs7yO14dOVjwxSIFRR3dG54u17l3HZqHsACPpuJhzaRPqqv8c/cYepExutPwT9L8Hufhh3xX+YSkZ732UMlzLSCsaIhVF8fMGzsjK4Gz5pRvcixOp6lLDvVmaPc48vZXj+FL505ypdMIdRPESfZXXUdCfervehg7ooKMDGWfoDnAug0vli4kKM2F8KvBPYKYR4Ntr2/2itf3AB2n5RIwS4q76FvehBwtG1CHsaq20Xwsoa+ddd70fnkpm0A8oYp/RV/4BsOoys7zM1RPMRHHN0feRRxmi6o9i9P8bpehKtJOHQZsLh9WZ71+PIhlMm3vmcdN1jrumM4675olFlzLYRDFyDvfDJohGcmliCd+DX0dPdgEa27iG1+qtzrsOpJhdFRr2w1quD9prxT748MuqFmZYpgtM3Yy98Cp1tLjk3Z9hs42cucktY6MmlaNtDTXWYiKdZyY3OL1BGapHffTbpglpCL3PnVjfqdtej8ftyLWnwnvsvpsZpwffUP/5qZNPxRM1xDpy3YddaP8T8m/ddVsj6AWT9QNE2NbYCnW2ibOStJEHfTbgrv0tq07/gH7krWoQ1hToqj6orYZKQUpv+D3pqMcIZR4UO2V0fQE12gU4DAcHJV+CuvYeg/4ZotFrLV6LSgmoI1jTumnvIHnin8eerFKHM4h9/FektH0XWD6CzTWR2fTDKsIxuf3gDmV3vJ33V388p5E5NLCXWtaFTBMdfE79PWYRDV6CyCyooNsbco0qRPfD26PPQmJ/guTyrOJ36amiQk2D7UbhkLeREz2bplz1mErxiQ0DNcxML9ptw0WpXm1gaX+pPOfh9NyeGfQ4kmacvANTkIrJHX0322J2oqW4AtN9c4TdnozzzwxVWFnfVt6l/yZ+Q3vKxKlPhWXy1fhPetj8ie+DteLv/LzJP/jlqYklk1M010S7Zg2/H3fBpI11wTsxEb4j0MP6pW00ZwJw7SbkQpskeehsA/ukbKatWpG10pi2qoTqHK7sjVTI1JbE/FWG0eoQ1zZzWC1QKM6uZ6wJ3bl0igMbDiAX7aryuCcV0lt43x7KKAYhZBLakh9P7E0TDyQoHCBA+TvuOWSWSCdMVPgMZSVgn1EqiFTPPKS2MHJy8DbvnJ9gLny43agBotNeK8hYgU6Mzm+2pKsqLsxEZNZWTB64w5Q7T+MdfidXxlBEr0xSM4uYSLSPQk0uIH5FK1NhKvCN3mdJyFRKFdKYd5qBZIlv2I+xJdFzCUr5fpZs0Vsezpgj4qVvnMBuqJeGncgISgL3gCM6ye8k8+2H01JJZ2jTyE/7R11FeBaoaDohMlXV2P3LDPYmwfLKHeuLb17ZJcJsF2Xg0PlxXeljt2+fQ74QX7Ii9oTNeyvNyQk11xRdGPvEqUDZW1+MFBalzGKOY2f5hdGBG1FoLIwesc9P/QnKLqXOh8vHhidsJ+28CLZDNh7F77ke07gR88yczZuYg/Ji+FFJtcVcQnrwVsu3xbWgLWXEECVpLwqENBKdvMEWhMfHpqc0fQzYdrXLdguIiIoyKZ4wg68/grPo6MwXBc30KmFG2rJVafOm28fn7zaS2/DPxGjwxyFzR8hqxx8HKUtz/SHyu7hR2z32kr/xHhJXFWvgksukwsfcq/Vmeq0HYHs6K7+TzM8y5HqJuEHvhk7X3O2H+j9gbOnuZHCgvdZbbljPwcce80AmHNhXoaBegBeHQFbgrv0ngjpgiwkXvaMuMIvuvx1nyC/wTt5nRbVmlo8L/vxDLJFH0RvSbVOPLEakh9Oh6U+hB2wh3jNQV/4bWGm/bH1FV+CtvFAu/prm+5raVLtqZ/6+UZKQybXg7P2RkYLWZHcjW3aTWfQGZGiG95X+DXF8hMaywnwEyPXMNp+tJ7Pad+KduQY2uRlsZnM5tiLp+U0w8s7DKfZ5DspYMCMeWY3fsILX+81HBcol5LhX878pBtuwzpflmXeA26pdq6KqSdoQpoNH9KM7ih2a2Ck1q079EM4guiuLtlYu37zdweu/HXvRg1bUPp/tRZMNJgr6Xov1GrPad2AufQlTK6UiIZd4a9sIReSXjDpUN/GwGv3TEP5cXQ7X+XFhUhXBlbaJUBEbzRGZBlehVqxRqYhkAwYnbKa9rmvt1xY3iY64Xu9A5ixFSTkEN0+gsr43sgbuxOp7BWfpDgrNXoid6iE1dlyEi3Y/OLJy5ZOyLoDSyIyQYuAZnUXEUhtaQee69kG2m8CWnhjcQ9N2UN1SibhCdk6ithC5f0BN2BnfpfUCxXEFq5bfx9r67il79OUS9aGHqjAJW227SV/0d/umbCAevjmQd4s6xsJd/j+DQ2020lMrJOscvYquB6+LbUS4601l+F0KT3vJP+KdeTnD6pcXyEkEz/tG70GEKt/f+qrdmNR3DajpW9ZiE6sxLV0ycm+VcXS+/DJfNxZodWB074rMgBVjtO8z/pgeJ9/9mEfV9+P3XlRv9MnJl33TJv8MoPt4rWEQLgSw4oxQn11RY9Cp1qWgbNbYS//AbjPbLxBJk+/bi6TeA9JCtu0lf9Y+kNn0cd+W3cVZ/veSaFdCpKPyxYFOQMgUvMgspj7pIEfTflP+nsDKIhlPIpkPMuFdi7q2GhUg13Y7KNiGbDxXcY1x7c1xE1QKRnomUknVnkenBgjq5Me3JLEwtJr3lf5Pe/DGs7ofj287PiCziTUSIqI8vBi4sH7f3fmRqqLwPKmVE1eJmoQkXlHk1Yp/NCBfur3Uknts2mzHOHXPxRuOzI9NncZZ/1xSLyEULaIGz8lv5hVHZdBRRN4Ce6i5YuFIgQ5yux8js+ANmNxoWsnUXaImeWoRo6MPp/bHRnhGYsL6BawiGrzDRI2GacHALMz/6SmFxleLmRdEMQp29puB8BVYGZ+kPsBc9ghBmBEfTMbTGFFwIZ/naSq+sXGD2yBuiTNwKz6JkYVlYGVJX/BvTj/0N8f5+lX+5xqE1ZA+91cxYRGguKz1ky27U8Ob4RcJZyRldAdols/0PqLv2b/NJWUH/jVVmBYDQCNdIMsjGE+jDb6D67Cf379LZhCQcXofTXVkSSk0vrLBHGskBt0x1JOECMi8M+7mMqqsZ4IbOpQBMDhyLPb7S9c5ndH8hff1FLzAewWp7jnDoCoB8YeQcQkB608fxDr4NNbTJLFrWj9Ow/lECZyJSF5wF6WF3PVFUaMM8r2h0bCns7sexOrajvWYy2z9SEo0iQQQzv38tjQKhM2ZipmeNxCg2GmgL2XSsLLtVCHB67sM/+noqv6xCk8jV8Wx+S0NnL1OPbq0YQYPwsTq2xTSVpvKkVmB3VTZs4eDVhAPRNfPXdVFTixDpgXPQpafkeAF+M+HgVdhd0cJi1ZeFQlhTyObnZ7ZMLTqH65p/q+HN6CAVSTWYBfqg/zqC07eYwuYVfeIK7HMNh02olUti2KXjxo6+52pYC0fZM9vKjXo14q59Lka6sJ3C8+Z6T5MDx4vayp/fc7xif4Q9TWrd58nuv5vw7JWoTJrx7bdj1V+P1ThCON4Rc1Zkha0sTms/zWtHmSoQ3iucvSivhez+d6DGl+UXHMubsxFNh7Fa9kFQZzJk686Yl0CQjozbbMJVEcpBjazHaiq+34bOXibULwgGtpqC3XmDGblLpMJq24m74nv5UWz+84gNDTXPwWocY8GV+5keKdnlTFTpa2WfeENnL8N7bosZPUvItuBc8TGyz32A6nVQa8EiGF+WN+xW59MExztikqU0pM+Q2vipopelTA+gJhrP8doaHc4Y9uzBtxn/fv6ec3LEBfcnPeyenyLk5VnObj7N9i/5iP18H8S5jLJLXwbn03Zp/2ttuxZma6tw38j25YRDm4tGiOFEC1bToCm1VxifLXyjVW5P07DyKKmefQgZv4Ba376Msz96V0mWa9yxIbK+D3fpj9FBCr/vFsIjb0Ckz5hixF4bwplEjaxh1q9dJJ8bR2PXYtjyT/h9NxMOXo2wPOzuh7E6thVFW5Q+N9myP5LuLY4MspoHaX35PQirPAxQCG3C/WILKRsdk2B0DcHJV6CzzciWfSy4cjcwEV/AI7qmlJC+8qMmfT6oixaFcwZxLrotmnS7nzcoVutughO3YSJehHmOQuOu+jL2wvIZibPsR3h73lMla3bmOmV9cSbzbh2VaScc3FrizrJNGKY9DX4zOOM4S+7HXvzgLNcqZ75Hvs3H0OtLbthrM7BzG4XP5VpznyUU96UWF0/uvFr6X2t/So9TA7eWjxC1RTjRTvN132dq//WE4+3IhhHsxd+leV35dDju2tn+FeignmI/c4zfVSicRQ+iQ5fM9j9Eey3ktMi19LAXP0Dr1l1MHTjD5J6bjLsmP/ovMRpCYMe5RiIau7ug+wBwAKCmGZK78ptmvSF0i2YPWglUpgGrYSx2xEpG0T0AACAASURBVGW1P0t45jpK/exW0xBy8g6m9834tcP+NoZ/di2pLf8TWp6A8VeXa6fLENFwCiEU6a1/Rcq+jqkD15I9varElRLFw1ue6XPZgqjZX7fcFPdWmVa8nb9n5Bdyx2mg7jRqYjmBVGBNItMjaL8J4Y5gtezHXfsF/MNvQnut5mWlBcUvv2ixV+jIpWaybVOrvp4f/Zu6rgFl5kS7iPoDWAuOGqXGhv7omduEZzejMx3Udfv49i+qFqWuNBueL8y3tTm4VK4Y263Z0C1aVyj1WbuBz7WfM8Q5Ss+dy0ujtK2Z7bMt+safN1eqXUdXXFDU2C1naH35V+Z8PeXVMX1sfYXCDiVGRmoaljQwtnu1EXHSJYJZp25DeYeoX/M0bvfzeCfX4o1PoAa3or02Ckeqsm6ChsWtTA3OvHziX8q1fYfARI2kNn0M79kPF/VfTbQx8tDbaHvVp8t8+kBUy3W9GVmrFEgfIRVNW+5n5NG3Fj8bbaF9h+DkK3CW/YBw8BoTFhi6ZvQsNak1X8obscaFPUAf/tCSGP+4KVforv8MQvh4+34b/EZmdPI1Tdf/B9L1mBw4jn/yLdEsodAo2zDVQzDVA323kH8ZixBQyObDpNZ/FnvrX6OVhco24u34A5NhHLpmxC1NH6zMS/AHe5CNw8iOe5GNM4vTouJCqEaPbiAYNTOl4PRNyKbDxrcfmopd/qkAK30LLS/7CtOjB2r6LC8W881Yz4VLOmKfq8FrbqtnbChemzmurfLR9dLY42YzDpXanst5c3lxnEu7bvchMke2lBkImZ5CpifKzp3tGpmTqxl/+jXRCG42BEJIvJPrYHJ1bGSGkCH+0CJSi57HbhrGXv84zplljPY1ULooqDINZM8sp6HLRJd4p9Yy/MA1qGwdqe7nqV/7BE09nTEv6fioqfxUfvBGY/eKBocS7afwB5biLjwaHTtcdF5D1xeYPraBYGgxVtMwdct2Ek43IYSKSTGw0RObENZ3ab/963gn15LtX45Ij1PXuw+reQJT/73wnArPWIPVeJzGRQtRyz5P5tgG/KEl2I1D1K3YgUxP5e9Tja0g/ucco94YLWarsVV4B36d9IZ/R8gQKz1K3TV/SzCwFTW+lFTnNHXLnkOmPOCBknZm7kF3ZMke8tDTpWsGudlYTpIihRpda24sNwMKXMLJZiaeu4Xmq437rfCz0xp0dgENCxcyOVBu+CsNduIMcpyhriXfpdI5tfalWp9qJdd2rcr0l8SwW3Z8pEFz24zQTyUDDrW9EKq9BGrZX8u1Ks0GZutfLfsrGd7Sc3P/VusfI3t6FSqbLhghKpqu+RGENlPHN+APLMXr1DStf76qgVfZlDHqNZRgy6EDG6l7kHUTxIU5ai2QqeLn7Q93xft3Q4dguJtU1xGm9tzI1KFrI3cETB++imzfRurf8pOCZ1DbOkQ41Ryrl6O1QE3PxKRbbvGLSdg+9St3wModBedIdAWtHqt+LLquItWzF5VpYGr/DWQObUU4Hg0bHqZuxUwEUmrJPjKHN1P+cxTY4c3AfqSTpX7VdlgVr5liQl4rFVapgLZRwxvRQTpfx1XYHs6ih2nYYsJLJ08PgOVFswtDqetRCKhf/RSTO1/B7OsCcZnUNtmTa+Hq+4razQ4uYXzbnahMIxlANh4hte4L+RnCbJ973Mu90ii8UoLjuVCpjXN9sZwLl2zEXmhYS10msxnkuGMKXwql/P/svXmcHEd5//+u7p57d/bWStpdSavb1oFPIdmywcYGHAyYIxgIgS+5CJAfIRcESMKRCxJIIF8gkAQwEEgMGIMxdjDg28b3KcuWZB0rabVa7T27c/ZR3z96eqamp3tmVjLI4cfn9ZJ2prvqqaqe7s9Tz1HVnmx/mer31i0Hlbzr5dW+IUbtYzMyD5KlXp9GMty6faTecCsL+1aSP9YHkWMkVj2BiJSYvvWtOMUk2FFKx23mn1rLkpfdQ2L5RKDMhWeHEBrIxSQvaBaxvhlSa7KMHd0ItvIACwcjVaRjfRQhqu3J2Si5vVaFtCvQTfREBqcUJ/fs+bVk7OjYxSjzu9fQeY7ZQAHWZ0rJ1VlKYybS8iksKTC6xxYxWNATC0R6jmJODdb2T7dIrH248jW//2yyz+yojFGWkiw8eQm5vduQZhy9bZrkhvspHD4TbP8qUI2FJy4lNrCv6s8uJMkdOMu1HtKTJNc8SiG3i8jArRSnN7PobZk1SHSsRk/WvnIwf3Az2ad3Is0oaCbm4E/pPGsvuckqAanBfT05j2aUcKxmi+Fag51rZ+5nr6m5N5z5YUpP/yFdL7mmpS2Z64nVc88uLimiVXJtJZHiVFa8LwanZ8auuzefR+4quaR73B0EVVJM9yTITOXryrQKvyUQTvDh8PqjKgl/v2pkBozDqxd0rFJeHWeLikcdU+fSMcAlqbE9GRZ2XYSTb6vmkjs60oHJ289n8E03BT8gLblffHA0Zh/dyLJX3U7vxQ8zdfe55eOCWHeOVa97iIIv8SQ5PIp23wtwbHWGL919Rwb2ojtnIQwH6d/jytbJH+2n85xnAq2bIGWV7k7irD7K3OMbMee1MokCeonUqjE6VreTnQjeXybMukmf/0MyD12BObXSTeETku7tj6N1uw+rlJDbu71ecTkGTt5d9m/NLiPz0CsQwkYGzHalFXGDu4kF7IVOZu54s7sDpaNjTg1QGNlM54XXEVk9xkLhIfJ7X8hiUiiFUUJL1JJ64cgGFnZdUs3ssQ2soy8jn0qTWh88I40uGSHqWATnMtWMyGtZ6YRFbGBPTan8oa31+yRJHSffhjk1QLQ3fJO3RmjmQm0lA2cxsZ2TQZhCWgxOe1aMSqoeuanEHXTM+x5GqmHyPUJViVCVu6QnyYmpWtL12hhc17Mo5ZKZygeWaeb+CRpXI7dR9QdfUTOmzFSeVN8KZsbPCFwgZBeiZJ88m+yhXhxRIr3pWZLDowgBiaHjITO/RnnoGtZcmvlnhunYvJ/U8DEiVj9a1CLe477NKErtb6EZNu0b9zP3WK2lI6VOV+85ZOdzwZugCQejvXVrqNoeLHv1rcw9sZ7sgSGkzJMYfpye8+YqcmoXtQW7vVS0DzyCnXsKuxCjZO5GaE7VQnAEk6UWZrB2BKmFkbFAixRI9a1g/NEd7gzaU4JSB1tn/rHL6L7066Q23oc5thlrPok/rdOTVc1mckC3aXvBTyqz8AqpPXNBfbqmHSH/7HZ6LzgRqOiEbvPl2Bt5Z/4/kQhsdLL44yflPggHoZVcBaVb6IkFUpvurillL3QGL2wTEiffXndY/e2CYlDqsTCiVMcV7kdv5CZd/BqWZll1Xp+8vxN1JYJx+lwxTYgx3ZNgSU/tTPrEVK5y7MRULpgEAxQA1M+EveOqPD+p++Wp7XnEHaZgwhQS+KwRn7UQagGEWhW1FoQqI92dRBgh27TaOlMPr0Ja7sx1crKLtvEeenY8gR4v0X3hw0zfe67rjpECoTskVx/GzsUpHOsvE27tQystg+yzK+jYvJ+OvjiZaddyiBPcd+kIMk+tpZaEBNg6x+9ax/DrH2amd4H8eLpG0Qhdkt5SH0iru5YBylCLWqx86SEy07vr6hanU0w+fQ5W30/RIhbWfLJGgQQh3Z2EbshMZ4ii+qFd96KeymNnW3hJhKO7qzVVt45mklo7Stsyd0Oy4rF+gnzU9nwPia5hNMMm8arbGL3+IpycsgGX91e3iHZlkLYg0rVAdOhOIl3jZCdqScXJBy9ackwDaWsIw/XR+Ql+deIpjueXcgsvZYZO3sm/kse3n45wSK0+itb1BHa2CyM9SbT/QN06irZVWWZOWEirlqKko5FerVFSXm3rJz8V4ZlstYogSKEHEXjzxIb62XYQ/Kvig8pnJw6ftGVwWojdMLQ60l7Z494AI1PZymcPI1PujE+t482ugwja++4dW9mTYmQqy4mpRIWQVVlBhO5H2GzeI/dWEaZ4PHmt1Gl0zi+j7/zDjN16hu8BcXeN9EgdXFJeeHoNHVv2YbTlGdgxyfAFDzDx+BLycw7Lz5vFSbjzhcl9EY59/8WBi3C0iLuUXCVV73ONddadpJSJh1gGgvxx11XRe9mdnPjpdoonehCag6ZLBl62C623dplokOJrZOn4YzXZ0U4OXrsNx9KQOyLYpsGx717O0lfehhY1yR0aQEpIrjxGJB3uzstM5yoPY3biMF3bnmDqzvMapKNWx4wULrkLCVIjOTxGz4WPVORrUQvHrJcjNElHbxyhSTLkyi8LCVodLOi/4m70eCmQWDxSmetaoDRV//o8PVFE6PWBF48Y/23NxfzL3H/zKvsHADzGWXyG9yJVN5sm6TxnNyVzb6AcD07nCHNPbMDOieqqYd0kuWqMntUOmemTTyEOi1n5Y331JNvabNwf0/Pfg42shUZlFoPT6orxE7h6bGVvkpHJxoTrVw7qMVX2yt5kS3X8xK2WXdmTYmVPigf31htDrdRXz4cR9WJjB2o7qmyV3IcvncScmmDq8SUIXSLLEyOnFEwQheO9LN/m7i1QkjN0bJ2hg2qGYLonQXs3TP60RGmu9n2dwjBZcv5ozQ0dSLjlcdrtGgIRuI412uGOoWvAoOutD2HOx7CLEWLd2fLsLhko369E/C60MIz+aLOPNAVOyWDyth2YmaR73STMPriFzvOfJP2iY+FxETnE4PwJpBCM9h1Fi9jMPLgZaz6FiJZwCrHgLQ6k5jrmDRsktPUX6eiLV+IgPWcf5sR9q2uUtNBtOjeO0dFXdfloEbAD3mgnEKR74mSzpYbuq4HL9nLouvNqFL8wLJZf+kxo0DLVt4Jbe4f4ijnP7zx5I3ktzn+Zb/T9tgKJROgOqc7wRIDMdA4tYtN50VfJ7tmOOb4RYVi0n7mfwZ2Ng9zNMuvqCTp8tu8fXyvHGvXJH6dTFYl/ErLYVGk/ThuxB5F65VxvksF0nJHJXGW2firyPJkeVJlN6ymKxv27sq5fYTLU4yNT2cBZfxgaKYtmMlQr5sRUjg1XP0P+JSMsHGkn2lFk6qlejt0zWOe/dkwdzewEpgKkuvCslIv/dD93fGIjjqkhHdetklo/QseG45U+hAW9eztSzB5JkkjYrHrRBIfu7MMxqySiRWyGr6i9mXtWabh7wsTr5Hn9anY9gsYCIG1BYbLebwuC0nQbqvKSwOxDW1iyZRZ6qrTljXXD+H7+8fsfp39+EilhtK2PP7jsj9jz+jHS3a6CGLtjPZP3ryHQ/4wAy/1dJh4cJrFkvnJNl+zYz/xYjPyhQffFJY6OFrGZ3buUB/6ul8GLD7PsglE2XD7N0z+I1xAzwiE1NI0es0jHGqf6tq+aYvj1DzF2+waKU21EO3L0X7SPjvXjZKbry1u5OM74MI6l8X/P+0O+c9Gvk3pUMnV/L5i195hAUBpZh7G5Nm0zaAKgxQq0b72dVN+BhgkOQfGyzFS+5ZRmT0Yz624x2XOttOHPrAua4ae7k5UMucUQ/Gkh9va4USHvo5nwOPqFq7ubztxVwg6DVz+obCv1/XIALt7Qx8hkrlLf30c/2TdTIB5OTOWwSxrjuzrQDIfedaApz6cnR403QLAlUnO8ByhvI7B0aIbx+wew696oJpi8fy2bLyvROVQfc1ADwunleV7xqccYf6qDyeMO6eFZEj0FoHGQu/jsIDd+bdhdeOII4j15es86zuRjS5FSYMQthq/cR9eG6RpZYdaPh7XreyrH2rsSOJaGHm2cr+mNRUoQulNLhI3gCIzp1UD1JRuZqTxLUw5f+u8PkCpkK86H1XOjXPvDj/Cyd36VPOWc7zOehIeGq5k5IZCmwfy9yyrELjTJ2tc/RWluP/MHezn20zOxC67bpTRrcOjmNRRm4mx7yxhjTyeYfbbTjVUKiZEqceZv7iGarsZ+GinDtpVTrHvbvb5r5XP7TedYeHbQdTUBSMH43euZOe8gsa4cdkA8QNo6lNpaIt2wNOWwY0EJC147FfLv3kjGN9OunmvMBa1kz0G9pbCY1Oyw8+6xFc/v4GlE1zh/oIOeZIRl7THG5ouhBD+YjtcpAI9QB9PV2Zu/vncu7LhKyh4uXN1d104jXLi6u+6YP0bQquJY2Zvkzj0TTDzex75vb0TTQRcCIwKv/OAhlm3Ih8oKi0mEnWtfWmDT64/wxDdXUhcAtQXH7xik8y176+IQ/njGcH+SgbTN4zd3sfs7g0TbTda+ZJwlZ2YqdVQsHGvjyWtWY5eqhJYbT+JYGi/8yF04JQMjYSK0cEXlwbNGlvQkK2N0LMHxWzby1J1LcCyNZF+BVa/cQ9f64DRGqJJ715YjzDwx5BJPGZruIKWoz+cX0JmI1llPlz5+O7pTS2caEHFsXjn6AN8avtg9FrFpH55g/tnmW/d2n1jg21//Q37/NR9mqs2936IdBcypPqRdG8B2TJ2xnw1gveEEl/zZfg48qbFwtJ14V4GONTP09yVRH3mv/624qjykexJMH7GZfnyQzOE2coeX11h+0oGph4dZf/VuhKx3s4mIRef66dbSmkNSilu11FQ5KgbX9YTWDXKvBcpsUK5VBeCX17TNRcg97emOPclqAM4jVY98l7VXVwB6CkA9r5b3yF8l48F0nPMHOmrKH80UKt9VxaDWCTvmP9dMlh9+ZeK3JM5ILeOGbw/ilDQc3G2VisD3PraKj91wGJCB7TRSHqqiUcuNL82jxSycYm0AVDoac+PRQIvDUxjeuVJe4xt/vIb5qQh2yd3DZPLpTna8aZzzrpqq1PEI5MgPVuJYPiKTGvZCjJWRIUpLJ/Hexel3Y6kE6pG+n/wfvmaY0Ye6K26d7HiCp7+6ha3vfIS2wYWaOn6Lx3n9CM+UUszu7UFoEikFq8+f5+DD7dhObZ81HdZuz9Dtv0bPZIiX6p3bMbPISitTkyk1d0/QK+lqEaXIq7iB4eMH+fQPP8G73/UvlTEc+npHYOBZj0gWxuN0D2dZvcWBLXPlMwHWaotWpNcmQGY0wZ5/P8MNMlvBr1V0LMHcoU76zh5n4vEllXiOFrFJ9OY588VZNCNZvi/CrTL1elUyvcqJD0ExJfVcWFJBUCq1Sqhhacpq2bBMOX95NTvNLyesT/5U7MUoXRWnldinctWcJY/EPbJe1h6rIX1wlYBXZ2y+2NSVoyoGoKZ8kPJoJEfty1TOrCgZVcay9ljDPnnEGqYEHri5HenUP+xSCnbfk+TKq+zQ+lJC9mA7d92uk2hLs+TsSVJd9S878Pqwd8VMecZXCyPqsGVHMdw6UI5/90sJFiqkDiCwioL7vtnPr72xwLrlsYorbWVPiiMLicDXogkNcrMGl17gvkezkaXkD2h7yGd0Rh/swbF8cQNL48itK3nRHx1sSGRLl8ZZ+scH6HKm6RotEok5vPLVR3j4hh7u/no/svz6WaHD9l8/wdlnGYxMlmr6tGfNZoqxOIli7cNYisbYt2ZzTWD/senglEKBg0QjTp4uZng/nyDi2Gwc3csZYp62dasB6F9pkjlWv52uY2oke4o116dRnEq9JkGxG88i8mQ88tVh7IIaNA9WTslYhK1vG+Gpn8xy9K5lCGnQe9YJXvDKaTRD1ih9tf0wS00lP3/as3detd7U7LgwqOeCSDao7bDzQbGkoLL+dO06Wf5sq5NMqDgtxJ6IBC97XtYeCyR0Fd45j+TDyoeR52A6XiHisHY85RFUriNWNdXD5KjWhao4xuaLDRVJmx3H9s9ocbPf+o0Uy9qLFTmqAnEc+M7fLuXhO+OUCgIjIhFaN//nr8e5eEdVmXhtj80Xufy8Lo5fOc2T/9NDqeD+HrohSafh6rcVaW/Bmjn2aCdWKXjWePjpOOuW11oY51xc5NjuFGaxto5tCpatz7HviGSwK1GnVBpZOR4e2O9gRCUlvy6TAjmb5uINtS9fVmMjfvnx2Wru/7mvmmL1efPs+1kaZHmmPlgCklUZZdJ78ozzOTS0jjWH9xItub+NGYszum4T8xdcxMpySslQd5LSXPB90MMk69nHy/gR7+Zz9OBGKx3DYF3EwssJOeeqCfbdl/ZNmCUDZ2bZMBwDYkrAP9kwEaFi0SkEHlTGNgXTB2qDyUHQDMnQC6cQGmx+6TybXzqvtFMlKlXxeEQcRO5BCicsrdn76x9LM4IPczsGlV3M8bD+N8rOa1a3FZy2GbtHzH5S7IjpzBXtmu9hUM/NFW0291dnQcvaY6zuitfI8tptJqsRPHlq/4PqBrXTTGG9/ArJj66X5LK+WZgjOG9n7bXS8jku+OLf0Xvdf6MVimzmEt7tfJZ9rMcsufW//pF+vnT7caKxWqvDi20MfLTIXdvm+J9vtJPNaLz4cpP3vd8m1uk+IKpFFYT+pZJnK3t4VyEdwfByHdeZVFUIr32TyX3fs5k8ToXco3GH7VdmuP1zK3n2UfdhWra6xJs/dIJtZ9df1zCFnd9U5FsBSlHTJFvOdhoqelCUseMwaxZxlIj1WVsNztrqPWAG6mPjVzBf+bsvc9GN32Dbrd/D0DVGX/cm7rzyzayMxipKZH5axwlZN1Y0ktwhLsEwfdfeiDA+OFz5Or4viW5IbLN2N8W50QRD3Uk0n771K6Eg95yKoHLLUnE+q8t6dxrg5agLXbJ0xyjdaxbq3XkhiQZQn+1VE6dSSDpoJh7Ujmcl+qEmY6hjVJVfs/hOUJt++TUu0BCF2ciC9K/dWSzBCymDsoh/vth81tnyulvuqCFE77O+MInd5r7KzU/KjeAv2ypR+8v75bTSB6+uX9EE9SlMXkdMx3HgN349xgM/08nl3IcnkZS88W0l3v8Rd7ZuWXDNF6Jc9YkrOKd0H4ny7hw2gjk62cAeJnFnp23tkr/5/Dwvf2njMar9965/UP/BJfvIxDh9113LE4/FuPrWP6NgKgpHkwwO23zquycCc55zC4K7r0tz2//ESHdIXvfWPJ/5myQTYzqOXd2TPdkm+dzN47R3yBrXlmd5qK4wcIn5s3+b5PpvxCnkq3ISSfjSDbOsXONGQIPqecejP/wBL/mHD3HtVa8GJC/aM8Kuf/gsvQPVFzP7rS5PXliQPuicVYIPXLGqznIBWLm6wJPzw8TmZoiWijhCw4nFeOiv/5n7d1xeKffJ317O0T31yiqacHjvvx5j+dpSjVXnKRXV4vFbcWpf/eU8/PP7OrnvJ3FsxeVlRCUrzyiwYVuenk1T9K4Kd+V58JNrZfyL8PsHwW+dhJF7kMs0yKrxSD8MQUoqVFk2ScaoKetTOnfucfNhvvvb2x+WUp7XrP5zMmMXQrwc+AzuJsv/IaX8eKPyuhCsSNdujKQvTFY+R8f3YKd66GgLeldnfR27QTmPrJqVU8v7cbJKQx2jSpRh9TUNvn5tkRuu1/nutw2kA719krgheOrBCDsudHjHO2Icv/kpPlJ6oELqADqSOHl+l3/j7/lQta9xnQ6f1R+kUD1416gjplcUkTr+gaceYegNrwHbYlWxyD9FxvhT8Q+IVAzbFgwOOXzzOyXS/W3BM/52WPMHBf74fa7Mu27TWZhVSR1AYFuSR36U5o2/XaghUtUFp5L02HyRd38Qlix3+K9/j5OZ1dh0lskf/EWuQupQH1PxrJjUk4+x6aPvxSgUEOU0mFUP3s3SP/tdjnzvppqxqOSu/g1yv6lQg+2Xvy7HT76brLjBAGJxh9/9E5PdO+/B/Py/sur+O7EHh5h9x7twNr+AQWW88RB3piYEy9JRBtPueT+B+Qk96LoczRTq4kHe+b/4eIF3vclg7JDhLpZ1YN1Wkw98doZYHCBV/hcMry+nmnqswiPMC1d3V+Jcav/9SRUeGilfj1CD4nIng7AEj5o2A8bvZd556dWt4pSJXQihA58DLgeOAg8KIW6QUu4Oq6P7NjxSSd2DNrobBs6sIWO1nJ6dqvncnerBbuvlmp8dZt+JBQQCTYBuF9GtAkTiiOgUQoAQAt0soAnQBBBLIoCSI0kYGpoQCEHlb8mWCOHKE+VjyXI575j71/ssKNoOAkEqqiGAgi0pWE6lbCqqI4C2qE7ecipttUd11m+HnbbGP3/KwDkgsEzBN2+GrVsljz5W4OLuPTwu19EuF9yAnnTQpESTkgG5H0O6GSBGCs47v8Sx2Wq7gmp/O2IGM5ZdGVPdeICELkilDI4umAgpWf77v4WWq85q3ml+lrdEv8k9r/4EsXe9mReerSGExlyxqhg8UgxyRY0e1rADXmhfLGhMHDHqXF1zRbsmiK7i+EKRN7wd3vD2QmgZD/6+rLrmC2il2sR+zSyRePRhIgf3Q3/jxSk9yUhDV5sHjyA/8Q8Ofx83ue4bUYTmxjf+6ANFXv96CfQy9ed/zmH+PLAuwOt+o8RnPhZRrBMASU+P5OUXxJnOV5MMVASRelgbQUh3SL7+w3nuuMdh7JDBivUmq8+wGtZT++DPaGuWUBCGRspqc39bjVsWms+W/e2rlopX13/t/NdWldMoqaPZ2hxPCTTqXzOcsitGCLED+IiU8mXl7x8AkFL+fVidVatWyg9/6INIPYKwyw+frTyElvI5lqw/H1TOcB+q/bMWmWJ1hqaOz9sWVVYPIMu73Xl/kTJwifuv4EJdg+m9Sa6Srazr1fNCBL27x10wg7sCEcCRUCyi/CjVWrEYRCIB7wAS1X0K/b1zFZf7TRPll8jJ2v0NPSUqHAdh226/52bRSqY7rlR5vpM1QWg4XV3IWMw3jur4vHmKoNq4f7yOrJbRtGoZKcG2wTDcqo6sr+t90bXqd9uBsVGNbNYdpDfmFasc4gmvjKy0YzqSiC4wNK1yvFZ+tX+GrmE5tZsIR/TacVmOpGS7nY3pWk05tZ5Vbqtol/9aDrHyi3ZienAQ1tA1LLv6DHt1PXj1irYMlGEo/bFsp6ZtgJih1dULasPrh3fOO+bJDavbKrz+VNr09UuV65X9y/e+8xfmihkA1M2LjwIv9BcSQvwe8HsAF/j7DQAAIABJREFUy5cvc495ZF3MIc0iIhKg9YuuZpOmqx1FJOaSeJnUK/XK39e0AW2iUr6mD7665aPuH0NxDVnuzK1yWY0oUou45F/uD0a0su+K+rO6x2TNcSlxFZMewYnEKze7Lm0oldOk9AhS05HAXEbj8BEd795RH3QPwxwgSQ4hZFVhaTpT3WsQEYN0soiugRQ6OHZZttJhaSNLRdANpObmVMt8EaTEjsfd45RJqxzpk7aEsWNQM26XHWQkguzrrZyojN37KzQ3fUe9XmUlahc0TN+sXdMg6a0cFaL2WrpbTrp7ZUkJmla57g5qwWpbDoKqjpcIy0QqD6c0EqDHkUCHVgQEs0nXpSAtAVYJ5aet/sLKWH/hiIPhm8iNzgPzgaVPC0Tlv+qHSqKk8JfzKUXlnFTKC0R9Xa8MVasaCbaUNX3wrFKvZvXpqf4vPOu7PFFQ5xxBkwwQNRNIXXOt9p62aCDpe+Qd011LNEhxABg6ZEutxxlVPBfEHhwi9x+Q8t+AfwM4d9N6+daXbnePz45jTyxu03y9byC0jrHunBq55vgRIv1D6H0DOANnuvWzU8jZ8dpBdPZX+zo7TmHXfQCVuqKzv65OUL2g4/7zXj+00d2VcXhtANx8Wzuf/LuVZBb8vnhJZ7tJNm8QsfJ8ij/mrXydGEWK529H+/fPIzdsIDruvrTATrkr7GbnBHPHMqwcKFG+lyrXR+8bQDz4GNp7P0rlEbIsrHf/AdbHPoaer67atNt6iW3bhti1C6FaQokE9oc/iPP219ZcX9HZX+mDH6orrVAQ/N1nl/CVb/VgmoLXvHyWj/7ZcXq77br6ar1GCGrXc+tFPvZX6J/+LCJfzT2W8bj7NJslvvqW3wTgbd/6FvZVV2FNTmE8+yzmOeciPvQB5Eb//vHu5mpiwd0fRkqJmerBKX+u+zs2hvH+98HdP0MKsC6+COuD78fp7cFKdpEp2kgJDq5cR0pShoaE6rmyvHYrU/luxTuYL9kslBwcKUlGXIWXMISvD275rOmWlWUFG9c1pJTEDa0i3xub4xtH3nLImW473nmpXIuoLurOFUwHp3wuogkKlkPBcpASIpprrUhk+ZykaDk4ZZkF0ynLBKO8V5petnBMR1b6Z9quW7PSbySmXdsPDbe+6cjKeKB2fK5V6SoGyynXd6pygJrytpTYipWzujfFp177goZuSKjPPAvLEvTK/eV73xl6z6s4La6Yc9evkvd+/sOB58zxI2RH3QhwaqCPSP9Q3Xn/MfV4ELEDxDdvrxCNnp3C2vdIpa5KqiqpAzVt6X0DNW16xKh+98tT4VcMXnlz/Ajxzdsr8gpFjYFLX0QmW/sjpxIW1/zNLq75/nJ+fK9LXKuXz/OFP76LnVetqJCZR35zspffeU8XN/0oga47JGI2n3n/01x9xXi1/2096C95IyJXu6hGAnS043z4j5CXXohTjneIZ58lduklkMu6/gCh4Vz6Iuz/+FcwjMBr6ykyFdrRpxD/czvaNd+C2XnkeVtx3vVWGFgaeO1agXd9wxSKd120cy9FTNXvZiXjceSrr+Br6V6IGLxF2uhf+RoUCq7K0zRIJLB+cB1yyyb3GobEgOy23uCgfbFI/MwzYPyE6wYCpKHD0qWYD9yFXspgl+NFQfAH8rvNGfTsVKVOUKpwUAxLhdeWv/9B8I+pUdZYUJKAv43DmVKlrD+W0khuo4SIoGvgP9Ys202t5096CJJ/YKZQCaqv7ooHpm2HZdyp5B9UT213WWdbS66Y54LYDWAv8BJgFHgQeLOU8qmwOkHE7hEwQHZ0goXRCdoG+kgN9PmrN0SQIvBkeopCbctfL0hxeOW92bsH1Wrwy4xv3l5XJkimV1ZVLPbEKLc9uozX/9VLXBnl1ajveOUzfPwdDyIELOQNCiWdjsKzbt+6lmLccBfa938Elo182Yt444FPcMPDaygqi4iSMZMffPwWLtx8AnP8CPqDTxP/4vWIQsA+r4CMRjA//UG0Cy6o9NF6+gGcW24jmuxFbjkDNq4JvCbq9fQrRf7p8+jfuhlRKLu9NA1SCexvfwGWltMLO/r56T1tfP26Lmxb8ObLD/LynZOBaZTq9WsG/fxXIPy+H0AKgf3gD/nane7Og2//5KcRx2rlSYCdL8T5zy/W1y9bKY0gvn8z4oN/jcjWBs9kWwrnHz8KV1wG1FscnlJSSV9fmKw53gge+Tc71irUukHkGqQkmimYMFnN5KrEG9SGmr7rVyCq3Fay5rz2vDZXpKM1KcJ+2YtJ2Q6q58+sa5XYT9kVI6W0hBB/APwIN93xy41IHUBaZiC5+rEw2nwvMz/xB8n1FIUrbzf9286sOQeA0paXrKUSfHZ0ghTVWbs621YtDA/erD/oXI3MgT7siVGMsqvHszJ2Lj/Cnn95nBsf2kA+spTLzhtlVeQp5h5167T1D9GWsJjdPwFS0vM3X0E7MIoov1pGXnsjf23t5ns8AVRvjlzR4JP/vYVt/981AJROTBNznPC1hKaF+Mb1FNJa5XrMPvEEqXM3YnnXx6fg/MrTHD9SoxzJ5YleexOiWDVDheMg80W0r3wL83deA8CffHwD19wwSDbnbnb1gx+387rLx/mPjz6FELUWEtSSuvr7eO1Wym3ZiHhkV/1Y167CnnF3UpSFPIzX338CkI88UWlLVWR630BTxSIOjkA2ICMil0ccOIRTrq/Sg5wdr/g2tdlxdMXy9M77y6tWi+ca06HumBbgMlNdaRCuZLRRN/FN97W1GPgVU1jmW6AF5iNjVdGp9fSFybrUab8C0LNTLSurjoA2vcw8zNq6/ll+s1TpVss1w3OSxy6lvAm4qdXydsmsEqoCj8jnj9buxtfmI8Ugwk8N9FVkerN9tXytzN0NZVY/15Zz5dcStte2/5gqz9+Wv97sI4+QClBIxsxRrlpztE5+dnSiRhFFRo6jHTiKUNbTa5bJckZ5DdfzLa5WpAoOjVYjbvrFZyG+/dPA/gEIKdFOuNdOVWKN0Expi8NjoBtArX9RWBby/kfgd17D7kOdfPn6AfLF6g2ezRt855Yl/O7lj3D+xslaJeuzFFQLQVUset8AzvveiXj7n0Cx5CoUTYOIgfmuqzHHjyBL0o0a6xrY/q0dga6Omnb8Vooa26kb+0AvIpWsJ/d4DLtNUNx1n6uEFFlB4/ITmKpo/FaS5cmYGMVYdw5ydrx6rHxcvV5SkSk6+xu2FdYftVwYRXkKy2vDr3RUX0IjmvP2WA3qQ0WBASpVN3LXybLyBOrcYiqJ+/upKk4P3pqcyj6wZr0yU9Fo7c5iyP60bCngmFYgOfsJ3YNXtm2gL4Ck68v5P/vLq99V0g2SO390hvbB6mvC/ITtH4fXv9kRd+vazpXpyjn/rF0lSf/M3n9OVVbq9QAwRk8gnHoCameB7dxXQ+yGZrF96Jmq7EwO58qdtN94N1h23cxdRnSsTavrCN2zYILIy99/1QVmjh9ByDwR/5J5XEIpJaNY40e4+bYB7AArtlDU+eGtac7qerRyzO/GU6+zp4y8Y/bEKPQkcD7/YfRrb0LsO4S9tAvzlRfh9Kn7gwjMF59L5I5HEcVqfruMRbFef3k17dBnGYCrqN3fa6LeUtu8gmhXB7JYRFhVH7vsSuO8cCtMHqtRQn7ZEExyajzJ+22CoMY/VGsmzGWothVUxp4YRSc4WUBts84VF9SWYoGEWUJ+l5vWxP0VFtfylJzf8vHXU8evJl1oPsXjjdOL8anl1T76rS/VKqqxUJpYGM1w2rfthSqhegTaCnH76zaS2+icv4xK4ipB+8t4pOon96A6XhvNXEttitXhIUxZqd8dW9ImNAS15F7UYhyRQ5UpmC5sktESb994LeMPVG+iMS1Gz9WX0X/no0SOTlQyXqSmIaNR5rasrSGo8Qd2V9pvG3D7YIxOkH50D/GpOdhxLmxbS3bWXSjlJ/+FfBF99QCx/Ucr5AYgIwa5F5+DOTpBtDCFLmz8t2hEs4mVpt3rNHIcYVrIeLQmd04lNk8pQq2LTQ72k3/LZTRC6c0vhUKJyM+edHPPbJvshVvhvDVEAkjOUyKNfmdzagz++f1on/4yxoNPgwDn4vOxfv+NmJPH6hRTUBtQtRR2Hezig5/bxAPPXkRfOsufvOpuXj380zpyV2NMqpwwxeydD7O+1HoeufsRpPRU+Z6MZopBtQyCXGCmZ+W0ojwUK89TOFKR55fhETNQa+WEQFViliIzzEXnVxT+Wb++MIl87MdoLYxNxWnZK+bMrg75zUu3B55rH+yqIVuVaIHQ2bC/fBgpB9VZTFve96ELVwaSuge1ba+uX2k0ajsMfiXYNtAHtk3vJ76GlslVSRlwohE+u/l9/MfR32Iy28GOoV2854XfYaij3uppH+wCKemdmCF552NouQILg0uY2bYJuy1R59pS+7I0P8+SnzyIsGyElMiIgZOMs3vneSTX9FeC4KqrTJRMlvzkAZIHj4EQOBGDyRefQ3adu8JzJt/GRV/5HAWrdm1DPFLivj/8S1b/4Fra9rmEYXenybzuEszV9Te/Z0Gpyli1ilRLyOvj94uuq+XVMXcvc5EroM8tYHenmZ+cCw3qZ0cnGLt/b/3voyDIIvP3FwiOAyn1I/1DPPlInvd86Bw+ZH2Mc3mEEVbycf2DbLxM472X3hI4zqB+ByULNHK5qTLUeEpYtprXh7D+VAjeF79S64clNISNJSiDLQytJEc06ksjhCYP+OC3UDxFl7/7+5Xjkf4h4pf/1i8mK+ZkoBJ7I5JWz4chqJ5XJzNaXamRHmhv2A40JvawthtZGX5y9yutxcJ/rVSy0mYytH31JuLj00gkxbYUI+dvodBRfZdnkDXiVzqquyeor34Zc4dmufDJRzB8WSaOJphYu5KxzesbytCKJY7fc4jEul4QoubczQ+t4f0PfKiystVB53Ov+g9+49G/J3LkOEJZuScjBlN/eDVzSpyhFeXpV5QAt7S7yuWl843fMdk4jhPcXli8pRVrTkVqoI+PfHSYj+77fRLk0crXKEuS94h/4S8+9hTJaNXdpbrvwiyCSnC84kpaXEZakEUAwUosTNGopNnIxRYGv2Xil+WX75cb1H8/TobY/WhFgQS1+7wm9g2pNvnFM7a0VNYjZ4+YF1PHQ1DdZgTfjNSbyWvFWmjVovArKlVJeeTuPbi5AyfIHMlgGpGGY1Tl+sk9zCryFIFaznzmOGc/swsjwMdfaEvyzEt3BrbtyThyz0hg+x7yVowHJs7COjzL22b/m+7cJELK+liAEGQ2rWby0vB7vlWX392r3Rz1nQeqyV2NLCpPjv/3DLv+QRMI9Vwzheph9HNj7LTvrHuz6Al6ueslr+asTQs18lUZjUi7EakHxXpUmX6LpJHCUtsIsmTCrJswgg+qC1WrIiiOFaZgmvWjmZJplA3n79di0P6WD//idnc8VQSRt5+cM6PzdQTtL7MYBJGq14Z6zt+u3wrwvnvlOlem62SrctWH3SvbivJS282MztO5LMnCfneRTR2xlPfNCVNM/j7OHppjoM0hMjLKiZzGQrLqnfXKGZbF+tHDLHnCbXN2oJ/c5vXkdb1mFaqKYsCr2zyZzZSmd60SRpGXx37EutEHAl6NXIWQkuh0VaYwLZIjxxGWRX7FUuxkNROokdU0O5LBWm7XHQtyranng6zDsDonExcKUrZnOY8FXpM08yzXxpg/atXcc2oiQFDWlj8oryKMoMNkqhaMX5E2czuGKRZ/Hxq5tcYfcDPaOs85p+5cozhWEPxKyh1zvVJrVV4FLVohJ4PTQux2yQ4k5WZE7SdZlVib1WsVi+lX0HG/UmilD944/GWClETEsTiveIKex2YByD8TY8+qNWTaWlMKfkQskxfs2V15V+cAMGUkeKBrBW2D5bQ+KTl7z1MkigW0Mol3jxwjcXyKh858AQvJFO3ZhRqSsTWNvZFOxu4dDVV4ALptU9h7gmLUzbXP75tgxdgo6+YzFPfGmOroZM3o4Sbv7AEHyEajzB+dITUxzeqflbNmpEv6+5cPMtu/vIkUF0bcDQWqBBTk3gv6rlpTYXX88Rr1mB9B18wjaysRgaBNAgUkUyVmRhZqykNjxaF+DrMq2ge7agguLFEgCH5rpFEWm0eeqjXqz2YLswpq+1DNAgpSXJ78IGUSRtBBiiko1qZeK7XdRmQelh23WNfY82LGfljZL2FFC1uf/jxxKn1ZrJIJmqEHWSYVSMmO6RHSVrHiU00VC2zdu5vbeteS16MtWTWqhbDh0AGShXwNKXeXsmydO8ZjCNKDaXrmZomVihVSB9CkJFYq0T03w33J5WwvjpCyS0gEmpTsj3czFktX2vP3Q0iHLZkxhvJzSAESwb69PWzKT6PbNlp5bF3zc01JHdyFQ10jYxyNtDO8fx+67xVFw0ePcCxvMBdp/A7JVqymxVqKnuJuJHsxbj+vfGRpN20H5ohSjSsURYyZ1cvc3Hxf+TDlEZbm22q/GlmGfpmNEgiC+tCob83k+Ov7r0FQqrM/xtRqUkOjvoSlV6sISidW+xEmOwynhdhLjqwhUBWHc2YgoTZy0/w8cCRf3z+1X+1mgQ6rQE6PMB1JErjOPQTeWNRx+Mft3YDezC890I44dII2u0rqHjQpObs4wePRXjKL2E9t4egc3XMzdea8AQwV5kgXczxor8IozKEH+NB1x6EtlyO2YpA7NIO0VSTmmMxFEpS0xrfWlswYg/k59Mr2iJKNC+7NrPan0VWVynkB6NJhw/5ny7s/1kJDMpSbYa6jMbHrB8eJnTWP7jikvv9TRpYNkC/vg9uKhRh2XiX3MKvOK+c/HjYByBDFbu/ljNwEorwB1URvJ/s7+pEhM/1W0Cg2sxgF5BFpmEvGT7SNYg9BMoMsqrBxeOfVco0UhEqmzWIiQbKC+qNaIhCcqKAmRJwKnhczdj/CyN1fxsNiZtZhsv2KZigRXEaTkteUjtNXXHDT9KRkQTO4PjlEvkxmqnxP7opkhKWFDOsWJkicsJiOJHimfQkLyr6rXt+8h1q9MTOj8yyzTWxZ/6NpQHdmlhczx/FYO9+L9iMVRRN2fUYbvIhCAGlpcv7kQbJapIZEPdiaRj5e7r8QZCJxhIzSbeaRFJmJJGv64Y3RkA6/lp/D8Cuo0N6E99GPmGMRtDuHhvuWKT9UIu0vZDh39ij7y2/+WDI9Se/MNI+esZlsItnyhCKMiL2/jRREmIUVNrE5lOphZt0KoqaJaURw/C87bQF++a2SdyuxoTA3UBjxBdWH4CC+Wr5zZZrDZbdfUBtqXxspgVYshLA++NsNcrV58Rq1vZp410iGzpUzTQP8zfC8IXZvhuwRqkqIKoJu/LDZf6sIayuozFmlGXqLCy4plV0THY7JzuwxvmwsYygRCezPyuw0m+aPV8hsaXGeJaUsd/asZsGo34c+yH3xtKVzdkj/3MdZsqQ4z1aiPB7rDimpXC+hMaHFWOIUA0lSAzqwSDuWu9UpVTJ1AAuNA7kIkYVp+qwiuVyBi0oT5RdISGwheLhrBdPR2nWQcWkvag/zIKUSdKzR8RKCJ0iG/9ZSsjUzVqNsNADp0Ld3P2O9q5r2U500NPKxn4zFGeTKqYzlWLZpgL+VyVJQm6qcoONhfQsjfX+g2UPQ7LpRjEpt06vbqkUVpIgW6wrz6jTLPFP750cYaZ9MX/x4XhC7n9RVVG5en9viZDGeK2JIB0vUz2xakbuxOEvER0s6sEYWiEkn0IWzMm6wcX48kDQ2zJ/g+tjShn3wHsiMFuV4vJ3+wnxdHzxEkGwpzdUQu/pQ++XfmljKa7NHiOAEzpi9fam9z16rR/UkI+19bJs+TK+ZxUQQxZeGKCXbpkf4yZINWJpeaduqvnOpBp4DRe1HmAJwyufUG9gGDhptHNMT7ChOoiPRcEn9iJFixKgqGO+aGJZF1/wc6XyWmFO/46MGLLXy3NiipefBT06NftdWoDs28f3HcIRARFOM5Kt9PZwzA58P1fo7GVL3y1mspeGhkespzD3lv2eD+t8sbhGGZunQQRZSUFAcwrPr/PBb30GyVLSqNMLwvCD2VnCqs/Juu8hL8sfpdYowD/tEgntTy8g18QX7ESFgU6gyDCRBm9+2SbPOLw7lGXEpByGvizySN2ssgBXJCI8mBlltTLIiO02btAJnp+3S5OW5UR6O9TChu66SsOs3pcf4z7ZVXJEdZZkMnrmrEEAJyAqdC2ZH0MrHYg3m4LHMDEcjbZw5P8EO5okjsQmyAODJSCdbzTkkAgOnRrFQruMAj0S6aJcWa60FbAQakkktxq2JfkpC55iR4IxShoh02B9pc0nd5xaKz06y4/gJHCFwZPgOlzEcLiyc4D7Zy0Cq9sdakay30FQSbHTfNptFe3XXl+a4pHCifDXcd/7cmBzguJGoKevviypfJWV/n1Ti9vofJMd/vBVl4ZcbVqbR+JvJ98eqWu1bo/abkbNaJuh6BpULs0D8WVJh6dGLwf8KYldnwUGzen9Zf5m4Y/Pa7BGiyqx0nczTs3CYa9uHQYi6mbZfhnf+KZHkfLlQtzfGFBGyovaoJyMv6zfX8jCJzpG8ybqYoNcpsiAizOrRSnvqeLwbaETrhLYO3rJwiE5Z229vv4nVVpaVVo6bkss5YtS6QvxjGo4HeZ/D4aCx1qq/BkHQkSSlxZtyR0hSqtTxbjyV3HMiwkPxXh6I97LULvDK3NG6G1QAC8Lg/oQbYLrfMemxi8xrEab0KulO6nHuSlTjF712gZ2FCfrtAlk0HhRtvFjOoQN6zXtx6yGAzaU5Oh2THwp3xWAYcYSRYBj85O6v02mXuKRwomyhVfZd5JW5Ub7cvhq7bHm20h9VftgMOKgPYccajUX93Kxt9XijpIognGoW3cmmTO/aV/+iFg9+hRpUv9mxIEtmMXheEXsQKQeVgcYE75ezwZxDK5vlHnSgA5t4bp59WuNMCRW36N1stPKkcIggsQAbwbf02ki22r4pdPZG2lln1rpQSghuFR1caU2xw5zHRiCQjIooX9X7yZcVRZB7B+CbWi+/Yx9HQ2JQS5IabibIztw4/5VeXXNt/NixcIwlLczWAUwERaCjRVXgIOiMaCRKZqAiUPubkhabS7M8GutmThih/VFfIzmvRUg7JmcXXQtmVE/wSLQLW6u2Vshlea11jGi5z53YXCqD0yjD2owgGbJydNgl5vRoSCkXh3MmmyMWW4vz5DSDg0aqQsBh5cOwsXzv1vdTssrKsj/S3BWxGGvXKxv0LPqPBVkJHhqR9HPRz0b1m7nM/KTbKhopvqD2WlU6jSyZk3E1welKd1RmSH6i8d88YaTWrJx3/kje5DyrUHmo/ejCBOqJXZVf21aET+UGOduZZ1gWmRQR7tfamRdGaH2A2+NLWLAczpZZJK6f+YdaF+3YvFDOYyArPvghWeQN9gRfNZbSCCNanE+JQV5oZ3hRefbpRycW47kipRBiSUqbM2WORregxFVCOrAnkmapudCwXx5KCA4YbchCEaOFrSsiSFYUM9xgt/Gb9nggyVrAYyJVub47ChO8oDTj7hAIDNh5tpWmOazFuS+xhAk9zpVk6jJwdBb/EmoT6HJcYg97uIWUvCw/xkori0DiIHAQXJ8a4rGS1nTi4kdMOqGWUTQgrbNVBBGPSureX/UZ8B9T6wTJ9yNoNn8qZN7ISmkFi3HhNHO3hJUNaqNZfa+samktFqdtxh5G2M3OtVxOSmLZDK9wcqSwKZWDe36MKW8XSkibbc48a2SeaSLcm0tzQtTPzvqTMe7Pa9zfUi8hKm1WmVmS2BSQRIB9xHlWS/J263hdvwxgvcwTlzYF0djhMSsMfmR0c7a5QFdAop+NINgT7yKNjY0IDcYCzKJzW3IZM1qUvGawyYKdcq5OGUjcAOZYeUTdWKy35ivnWrEICkJjpSyyThbqgrkSmEPndr0TgFw2y9n2TGDe+wqnwLLsYe6JLWGJUy/rZKADT5uCSSt4UjGUiLDRzDBkZZXrKZFIfi13jKm2VYH3bBjZH8mbPOTEWU+mLoahAQ9YEdqjbv6+kLImtTSIlFtB0ASpWV/V8bdKqidL6K1Y9WqfPIQpo6C4RNi5xaKRRdOKddGKiyoMzytXzHMFISVvtcdZIwtEyoE6A5d0PJq0EJzQ45BsY0gIZnIF3mONksAhisSmwHnWAjcnl3NY8VGHuYKO5E3apcVF9hzDssCEiHBXroP1huTywvGKK8h79LaQZ401GhqK1YG3W8d5XGtjVhjsEUnsBougnoh1c0Fxss7V86BowxECpGSFLNKFxaiIMSnc/k8S7vIAl0xv1LrYZUYYSri3yx16B1usLB3YRJCVMewhwU1GN+vtHFdKN5VLKHKaQQL3amnWynygopHAQ1o7xbL18Sp7OrTvAveFgDuLE+wTcToxAxVFieqLAxulVFrAUaJMlhV9Npfj5fY0a2WBPBp3ah0ck91sKs3VkbAAktKiyylBIhZKnkHKYo9IcFDEGZYFYuVrbSG4W0uTljavzRykHxNnXrA30s6x+BIOFuwaWZ7lGhY3alUBhNVXvzciXb+V0Ki9MDmNrAV1PP5rGfbcNgtun8y5RjGTxchZTBk/fimJfavMuumH5QdMfaBzaEgheDqS5sFYTyVw+ip7hhR25YLouIG/S/PHuaZtdWCAVUWXNHmPdYwoDgYwIEtssbJoVvBF1oAokuNESClBRQ8CWEmJFc40JQQmgi8YyyrEomIoEeFJ2Um7NNlamsPEzdDZJZL8UO8hJW1+zxqjC6sSXH1aJPgvfQmW0Pix1sUrnGCSzKKxS3f9fJXxC51PGwNcqucZtHNktAg/cVJ0S4vfsE7Qj1knq5XZugXsFknS0sYKsLBMBAuKBbOaQlO5FvCsSLBW1rrjSgieEEke09rZZmc4kzwosQobOEyUFZRftg0MUuQd1jGu13p5lz1GFNdNksKhxH4CAAAgAElEQVThSmeap4p2oD/cq9/Magi6v6QQXKP3s0Vm2S5yWGjcIVPMCIP3WqPKPS5ZZ86jWyUOGsuatNS8XT9adSE188373T2L7WMj92sryuC5brOR/EaK5+eN5z2xn4w5ebazEJh+ZyL4pr6EfVqCoXikLNN1X5whc4EXIyodFvJ5ZkTj2cob7RniStaNTnM/bhRJXujkpUYyII/cS/WLI4kieat1gn+KDNaUqVyfgsW1dPF9o4MeTGYxKlk6V1sT9GLWjG+jzLPTmeNOrYN7tTSdTokLqN3Iq4TgJi14oZMpNH7kpECkQMKQLPB/7PHQWEYzeMS305rlMb2dXyM462AXSTqkRRatzm8eBAEcKwejr7In6cHCQvA0CeLS4VJnhqe0FN+jm8vkHJtkHhtBVhg8KtpYJqeJQsXtNCSLvM0eJ0Kt7zuKZHNpjp+KDi6lVHcddCQDZpZdRaBBIDXw2gjBE6KNmUR1aftOZbGbhwiSYVmkR5ocydc/O/7sspN5tlrBqchrRM7qzLvVeouF6vv3ZAYdU9HKeBdL6qcag3jeE/vJIDy5kIoX+kjeZKks8Sp7ilUyfOanISm14KFdK/OL9uOawFER5Vq9jz+xjpJqkCOvAV1YnGPPs93JsASTEoJnrCS7U9WMnILQGCUGUrLFyfJie5YBSnXjiyK52JnjAidDBzZFNJ4kyUoKdOCQQedHWhcP661F5S93Zlsi2jAIXEV4JbNcZs9xndbDq53pyjYANoKHRRvvs0fLx5q35QAlNA6JOFIIPqkNYUiH7XaGl0m3v5qEQVniUmaIluMfBpK0NLlSTtc9IBGgJyRqYQGHRYxjRFkmS8SQFXeODmwvTTFMhs8by133WAA8AmuEI3mTKwKsPHDv715pMiUay1GTC5rhuSDNVn3jjeqH4bmS68UIPFJdTJae/1ijuElwUkbjNhY7xuc9sZ+M9n9Qa2edna+btdsIDgk3t7lTmryznALnEbI/wGcBB0W8Lj89CHl02hsQcxAcBPdpabJC5wmR4nw53/AHEUhe60xWZo8JJNvkAucsZLlR66JPWjjAo3obZzo5Lnbq/b0q2qguykngcAZ57tA6+InWuahNzbqlyYqAYOfJwLNQXu1Mcy9tbCVHHo0DIsFOmamZCYeNzMZVmhYaXzZq980xkLxc1q4e9ghdHbFHxkHwFkn5zxvAlBbhCyzjPGee1zhTdbP6JZhsllmeEG01dSPS4UVGkXQxwyojxoiR4nChdiWs+iyMEGMFhboAtgGMB7jrgnAqpH4yz+XJzmxbJfVTtTzUBYFBVs5zZdk0kul3WZ1s2897Yj8ZPCMSPCTa2CYXkOWUM4Cv6v2VmdJO202B82dUeP5VneossRXcqaV5pTNdQ6QmMI1BFzYG1eX2EjhGlOuMXjLlNMnb9E5eYGURhKe36dT7aQWuCX6VM11pebuVqbiCGiFoFn+JM8utWmdjFSUlK2WRF9gLnEGONHZDEmzUZli/kjhcQqZyg66Q9XZTmKyjxLhF7+KAiNfNjIdkEQvqCDEspz0om2cBnbhvzCawX8SZLbvsHOFmI/mXfsWQrHfyPKFV76tuafJu6xgRy3W5FRHMYPCvxvJKoNiPe/U0O6x5dMWFV0KwSySZFc/dY/1cu2me63ZP1RJopd1f5DVYbEA7DL+UxI4Q3GD0cq9Ms84pkBcau0WyJpd7kGLo4NVl8q9xpkjhcLfe0bDJB7V2+qXJdpkpP9AwImJ8Te+nX5Y423H9109qKQ4TxdRqqXBOGHzGGOAl9gxrpbsASivPJE3cGWKj21ddeh9l8TnaHnTgHHueZ/QUbdhMYtTtq/Mqe4ptcqHiemnlJRhBxN8sBVL9fVq1BooIbtc7eDZk0VkOPVBWo03FPEVg4y1G60UAr7UnaS8793aJFN/Veyv1soF7SbqyMr6rcbU9URNjiSPpxeRye4YbjZ46GQDzwuD/Gsu50p5ijSxQRONerZ07tM7A8r/MOF3KpxGeC+vhVPDLSexlTIookyGrBI8RYyiE3NUHP4bkpc4M92vtmI2CXkJwo9HDrbKTpbLErDCYLs/ejog4R7R4eN0yZoXBdUb55dRSsknmWOPkkUg2yxzRRbp6TgYCuEpOgTVViVXcpHVzv+7uWTHoFDlfLjTMe/fDT+oeWUqqyqqVLJqwXR69npjAAWLsDVhw5mFURJlHJ4LVVFlIXJfHAc1NN5wkwt16R8XV8XExRArHzVry3Rt7RaKyMZp6xkHwkBK3iEqHIVms60sEOEsucCPBxA4wJSJNF7E1Qqs+9p8nfl4B3P/tONXr8ly4Rf9X4m49XRdkbTRr66O1C5wTOge0RIXUTxaOEDyppbhR7+YsmSNN+CZVQRTruZSalQuCgUsscSRxJFc606x33HewbXKyLZN6WCmB64r6sLGK74muulCoQ3jdINeOA0xguHvkUOSv7MO8yJ6tbKtcW0HwJWMpUxgUEeQRgfu3e7hdS3OL3s0XjeVcZ/TV+q+FICv0QIXvCMG/GcuYLrdTQJBD4xv6klO+N54L/CJS7n6Fk4N/DcLJ4Jd6xt4I0yLCF41lvNqeZIUsUUKQRaM74DHXkcy3tOXVc48zZD4wN1o9EqaMDhJnJYXysnZ3HP4gYRCCfO8vdubYqyWxhEDK1n3lQZDAACUudWb5id7J47KN1zpTrCunGp4gwhDFGteTjetDToRQfg8WBlTcQ5c5s8yj80hAVs+0iPBJY5DllEhIh6vtE3SEWEMDmDzTwliDcEJE+UdjkH5MIlJyTETrfP4loXFYxFghi3U++0dbjO+cDJ4vM+TnSz+eT3gursn/b4kdYFTE+LwxUFmOvcbJ1+Vim7gLXPx7wfyi0OmYgd5aAYwSoRcrMPNlCoN/jywjJW0SOExjsFlmeaM9cVIqqkO6GRqPa21c4syekqkncG+8Fzlz5NG4R+/gG2IJV9lTnCfnWVZOz3TTFV2zMoPBARHjLJnF71zz9olREUVyqTMbSOxuJwTHiIEAxw5XUyefwFltZ5xoQ034Lb2Pd1ljRMqrnksIpjH4sb64923+Cr+Ch1NyxQgh/lEI8YwQ4gkhxPVCiP+VkRsvHW6/luB7Wg95NIoITGCvSPJNfclp6ddF9hwvlzOB2rdYTpUM4gsH168P7r7pkyLivqABAtaENocNPFuWNyki/EDraeguCUJQWc8SAHihneF8OU8EN43TWwFaQnN3fsTmBTJbOe6h/tUYVbQ3dLJUcb/WHjqW+8XJ7a63GEyLCB83Brle7+UWrYtv6kv4F2MgdOO2X+FXaIZTvXN+DGyWUm4F9gIfOPUunT70SJNhWSCDxggxrtH6+ZrRf1oesFVOgZc6Lqn7qbgIHBUxHtTamQhYFmQBd+n1G/QfFrFFz9ZtoIjGrcrs8X49zd/qQxwgFujA8PenUci3DRshJa8IUGA60F7eoiGBJEZt9o8XE8mG3MZHRMgbTHy4XetkPOA67iH+C/vtLaHxgOjjb/hL/tS+hy9Z/8PD9ttw5OlxAf4K/7txSnetlPIWKaU3aboPGGxU/vmMflniPdYo58gF+rFYS4G3OSfYWA4a/qKx3anfahZckrxX6+BL+lKWUqLPtwpSAqPEGAsgtRkR4VGRQt15PSzwWsQNSD4g2vm0McCczxW1oBn8p9EfuAZTJV4bl3jr13C6OE6E1TJ8W+VG2TLe51EiNWtrXReO4GY9/L2vSEmbtEhIGykEnzYGuY9UzRx/mCJ/YB0jcgrb47YKR+p807qW+5x3MsMaJtnAHc6f8T378z/3tn+FXz48l47j3wKuDTsphPg94PcA0qcpENkIV9jTNalp3iZdV9mTfFwMLWolZjMkpM1OJ8MmJ0tO6NytpXlaJNkg83RLk+MiSiLkHaRFBAdFHFsILrPql/EL3Bz9dmkFxgW+q/cyIuNcYGdow8YE4v+vvTMPj+us8vR7vnurSqslWbYl2U5iO85CAgEnJCEkYU1CYNKBBjrQYRvCMmFY0jND0w08PPM8M92zQQ80Tc8MgQToJD0DhMXp0JCFJgkQ4uzO7qyOF1mWbMtaLKmq7v3O/HFvlVWqW1JpLbv8vX+5SuV7T6lKv/vd853zO1gaOTyuboIoppekIVr1VnjvrYTFlExSM08hT36D10UGy4fC/qKAF1wKb/U6OV6TBgpWhwcsJ+R7XhcX2YN0ap7dEjUo9VXowDzeTnBFOEB7fFnaLg3cbDo5k7GSb2YGZTkB59qRin0MLRpwaTjIaTpGHuF+08pdpn1aJ84kntOLGWQdIYfLYgOa2K7n06evpFuemNXxHMc2Mwq7iNwJJBXLfllVN8ev+TJRBuCmSsdR1WuBawG6TWbee1ILzbqEWmKIxKsRy/gCXYwyavlc0EsrQVT1oXmOC6NuSENUuWLjCp0kD3kf6CfFx4M9bKzgcRMgtFcQdhXhQWnlQVOaO27VgA12nLfYIToIuMgeBKL29O943eUpCVWuDPpLBlgnxeIDp+oYt3nLuY5uLraDrNIcFiGD5QNhP72kizXfc6GJkBN0ghu9VXgo6zRLs4YIpR7lAMs04ONhX8mG83qd4OqwD5tw/mgTdpC32wNYhK0SlaBOiEc6/ixbJnWhvskOcZxm+f4s68t32rPJk1QFI/TaTXR7Ttgd1TOjsKvqRdP9XEQ+AlwGvFW1ijE5RyhjGBoTssGKVGUCVi3n2GilPLmUr9yrRDGEFAodJ/tw/8Is5wO2nx4tN/Yq4EPRb71aRsTnRJ2gk3xJbD2a5bJwPz/1DxuNrbfjvD8coG1au7XDnG4PcZvpYLtk+K7XzTXBblYUz6OsJ/KZsRzODU7XUzA19dSM8hY7VKzWKUSVxXCt311idXyOHSkrH/WILAwq0Vi8k1M26Shrgizf9Ndwph2hIcHl8USdoFtzFe8Yklgmvfg6TjClucoQ0iJ7qz6OwwHzr4q5FPgL4HJVrU0yeoG427SVeSDmgIekBSWyAv5w0McVwQAn2Ik5n+cUHU9cmU4VsRTR6v3nXidPSiMPSQvf9nrYKRm6NVfxipxFuE9ai/NSZ8MmPVRmW5CKny/QoXk+Gu6lvUpRF2AlAf812M5fB9v5WNBHx5SLh09UVjqER57Ie34EU9YSZony9QWLhcLxIRLUgr1uobGqlZCrgr0ljUorNJ9ozWCBCUyZvE/1UfeJauY36ATrNJtYaqpEF8TZ8Erzs4RWqRCfcU6U38zqWA7HfHPs3wIywB0S3fLep6pXzzuqGrDFtNKuARfqMAFRo8tT0swvTAefCPewJrZhtcCrwkPcoe3c482+unMIv2RlOh0G4WHTysNx2uSccJjLbaVtyEgcf2U6+IMpr4iphmR3k9Lnzw1HKr6uEoVKFkM0HCPpvaeBvXj8b68HRBjG413hfs7SkaL3Tq+k+b7XRYuGXBPuntEUzBBV3awmF9WsE7l1nqZjiaPmfuSt5N3h/qK4V/qcDEqX5uiXFHkt9/BRYHBaZ59ymuQA7/M+zC3hNxljOWDo4CXe6X8GT1wTj2N2zEvYVXXjQgVSc0S4zV/OXdpOJ3mG4kEVr7ajRVGHw5uql9iDPGRaq7L0nczvvWWcERwqs59N8kB5QpqKj9fZCf7IHqjYzp9H+LVp494ZzMqm43lp5KQpvvI2fr5A55SBHVNJei+TH1caPiLAanKcrmPca6L38DN/Bb/Wdnpi751CO3+IVH2rqUT7GoUgHo4brLxJ07JyCNukkedME/9dGmkjjUG51SznbXYw0f55QNLskRRvtEMln0kAHMRne5WllpNZbR7l38gbOMjxeAQsk95ZH8PhgGPYK6YSWTH0SqYo2KfbQ4m32yGwQWefkumVDDebFYzH/iHTORzeYw6L9IV2qOIgiwAYx3DfHFfqBX7uRc1ZhZRUDmEcw8+9w0ZUL0pDwtiOw0RNRRHTtQclvRMfuNAOlzw3LD7bTFOJR0tOzJRZT5URSuvZ82L4O38N90srw3jsw+cO0364CU2ECTGMiccDppUcpuR9BER3Xc9LA6Pi822/h15SBPHPnpVGrvV75lxFJQIdssOJumNeHNOWAtUwHv9hJ63LZ564mcxWr4UnTDNd5PhksCfR/6TQGNSheUbxWKbJboQWeIomNvsrqsqre6qcrofo0Gio9QvxdCGIOiC/6q/lLDvCas3TKykeMqX5+odMK2+0Q/iEifHkEB6RZhpQGjVkA9nE2aWV7jySNrCT2CZNnK2jiXc6wmGL3Z+azjLb4UPisdlfweYZzpETw7f8Ht4V7udkHUeBx6WZW7xOVARRxVflVq+TfnwmKhiCORxLjRP2GbjfW8aZQbm7doDwglS2h52JUIQBTSUKXNTYI/z7cFcxN13oME0SsodNC60acJYdIYvhMdPMWILId2iefxtEQ5hTKEFsuPVtv6coSOPi8btp9g5yYvimv4b3Bf2cknBp84DbveWMiUeDWv482ImHFi+MeWAPKZoJWTFlTW+JhlVUw73eMl4zJaVVaEx6jgZGxGeLt2xWlSlJHJRUVLpY2ICNL4LdmuOjQR8NcZGkAX5qOnm0ylGCDsdi4oR9BnZLhlvNcv7IHiCAot/K9X53xbmVU0lp9Mc/deW4kny8UVtKYSrS5Ff3JNgGFwTlg7Yfw2Fv8svsfm4wq8iI8qZwiBZCnpcGVmme5kk11x7KWnL8x+Bl7pVl3O51lMWYxJh4/MDvLttUDhBuN+3Fi0oW4Xovql0vODc+LC3cb1r4UNhflGQhSmPkEf55um7RSeyRDJvNct5pD2DjY0xguN7vnreYJzLpszaqfCLYUzaA/D12P70mQ/9inN/hmAVO2Ktgi7eMraaZ9ZolSzQ3tRpRX6U53hvuY21c+vasNHKzt5LRWPhG8PCKPZulVDMGrvBc0of4UdtPHopOiJv0UInPyuRjpIDX6zBdYZ7vVdlYY0X4jtfDGXqIM+wo43jcb1p5OR4ocl44xCU2mi1qgbtMG3eaDgzwxWAHzQn+8t/2etg/Tf19wYWzwDOmmbPsCMeTiytYLO0aLI6wT+IkHS8bqwjRhfKccKTi1COHY6lwwl4lE+Lx9KQqlZlo1JBPBXtomLSqO1nHuTro5W/8tagII+LzojRwoo6XfBAzjYyrlsnyVljRVyJFtBm8SnNVrzitCI9KC4+a0o7JTeEI77CDJWmSC200MrBf0mV3IxClUU7ScfZQWk1iVLnIDvJ6O0wDSh8pNnudvCQNfCzYQxf54h1IGuUDYT9/J6sXddVcaR/AIyqvdDhqjdvpWSTOtKNlq7rIrTBko44Xn/tHbxXPSSN5os3YCYT+uNZ9Pkw3nLkSFujS+ddMX2QPlm2YZmKL3labvAmcBtq13IT3I0Efb7ZDNMbWBT3kuSrcyxl6iE6Csk1tD+X8cLjsOCWxqKVFw+QJSzECXBIe4Cv5l/nP+e18JOijM/7dvCQNibX8WYSnTPUXf4djsXAr9kVipeYTO0wNsHySgE2I4ft+Ny0a0kzIPlJ0k+PqYE/JhaFgtjUfx5rJ4/IqzXodWICxbcsqrFrTKLtMJrEOMovw4pQB1G8MBhM3aH2U14fDiRc/D+ioMMawSUPeFw6wMa5wGcLnx95KtifMo12lOS60w8XP8FQdZ118tzUkPveYNi6ww8VS2CzCHknzhDQnntvhWErcin2R2GUyJfa4BRQSc8Cj4rFX0oQi8WSnHrZJIwHlHioFkQ+I7HWr7QNVYDi+NEydM5oHdkl6QfLTeytcHEbw2CUZtkljSS18DhggxVOTUl0ptVysQ4l3Hobozieprj+HlDRUFVHlE0EfG+O0VwpYQcBVYR8dU+5SfJSmeJrR5HOmsLwurrO/3VvOjd4qnpAmnpcG/snr5Fqvp+oN9TmjyqvtKJ/J7+bP8zu5LNhPs7r0j6MUt2JfJLZKMxcxWNLhmEfolTQvV9GVuEcyPCuNnKgTRQErrNYnS8dz0sgazVXl3SJAx6TXFcQ9BB6VFm7xFmbT7xdmOVdNGTGYQ7jVLAcRbvJWcY4d5lw7iofyiGnh92ZZiSh2a46Q8nb9Qtw7JcMzNHG2jhRXzQGRmdv9przkcK3mErtmPZTXhcP8ctKGZ1qTfB6jWNbbCV4rI4QIT0sjz/pLm3q5NBzk9Xr4TuE8HeaM4BBf99fMyR/IUZ84YV8k8mL4lr+Gt4cHOF3HCIkMxe7wOqruSjzLjk5rZesDG3WC/+X1cHXYRybeqK109KSKmCzCz0wnj8T11z2aZVM4SgrlcdPMi9Iw6y7Kl0wj19HNpfYA3ZpjIp5IdUg8iCtbtnhtbIntD7o1x5+GAxyvWQ7i8Wuvgz5JVUw7KfAH08rL0sBuTXNBOEwDlielibu8diaSavgJKna7rpySusmLIAkvDom8bk4IJ4pmxf9oVvK0tzTpl2YNuUCHS3offCJnytfZYX7jZqQ6YpywLyAptbxGD9Fjs+wxGR6VZn7sr+THczxeeUFgwjlRDkiKv/aP4yw7yuV2f8UPNanaJoXSHie9LwiHeJsdLN4hnBWOsIMGNnvLGTCz8z7Zbhr4Fcu5KuyjGcurGeMV4TgvS4bveYd7ADbZEa4I9xVLMVsJuTLcyy2mk12S5jjNlr0fBa4MB/iGv6bEJG06dks68UKRQ6KL1yTyGCZiF8nJdwyFi2bBwhfgSjvAfzENS7JaXq2Rb3+5A6dykk7gPCAdBVyOfYFYZXN8JdjBu8N9nK8jvDPcxxeCnXTbLBtsNBlpttxvWhLz9JMZia1uAzFs8ZZxndfNoTLH8crkEXZKhlYNuDQuUTREX4wMsJEJ/izs5WP52Y2IE1U+GO4lgxaHcWRQNuoEbw8PANE4wivCfWV3GWngHXaQG80qdiU49RT808+1I1XHc0BSPCFNJbn9kKip6cEEj529kuJRaSnucVQy4fVQTlsix+ph8RMvTiGQPPLccazivg0LgK+WT4e9JdOEUoCP5Zqwl2xsO7tDMtzgdTFRpZ/Ig6aVU3WckzUaueFRKoA5hC3SwueCXnqIUh6/NdGg6bO0fPUeEK0zC9ujeWCPpHlBGnitjiZWmQjRl+Qksrw7GOCHqa6qYu8hl2iXYIALdJjnbSObbLnXSwEfiydwp9fBB8M+ptatpFFO0THuonrr5B96K7nADnGeHSGN5Slp4g6vI/HzsAg3+yv5ia7AAP8p2F5xI9dfovkyeyVNf1w1NflzDRF+lzC83HHs4oR9ATjDjpKZJOoFCumFgsnXOp3gT8IBbvCrE0cV4Qa/i7U2y3qdIKOWDTpOF3n2S4qHpJnLJjUCNWF5sx2iRUKeMU0oEvtHHo4nclOPnn3ItHCPaUNFyKtMu8oX4DWM8UPVqnLu0zVZGeC94QATmGlfM4ZhRLz4VaXRWSJ73NmgIvzWa+e3s/DRVxFCLW+omswzS1i7fr3fzZVhPyfoBBYhQPiJ6UwcXu44dnHCvgCcaKvzeSzM/2xQW/WqHaLSyV3FjszDG2QfDPaWlfylUc7REbrC8jp6H2jF8tW4Fnsyz0jTjHk5IaoYyVWRT+4jzTiSaHlMHEcTNnGYhQKP0EJeDH2kGYhXqZPPGiAVB0wvOCK8RJoNUwyLFdiHV/a7XEwOicd3/B5aNKARyz5SZXNdHQ6XY18ADppU1Z2iSjTEeSHo1uRprCFCO+VdnNHPktveJ8Rwo7eKHNOv3KsVERXhBq+74rEKKZ4kt8p+UvzMX1F87nq/mx2SIR935o5h+LFZwe4lXKX+3F9JFoq/1chWGf5hlkOrF4pR8RmQtBN1RyJuxb4APGBaudAeLHE5mexcOJkxTLFJaL70SZrOBJ92D9hGIx2MlJ1JgL0VxrZtM038lRzPR4M+1k3ZtlWgl9S0fuOiSls8hDsrhl0mwy9sB5dq5a29ANhHii7yZBH+IK3c4S0vqWkfFY//46+mLV6l9pNa/EagKfRLmq/5x3GeHWa15tgtae4zyxhZwtW6w1Et7lu5ABzE419MBxfZweLot4N4NBKSItpILQ5+8FYs2Crr1147pwSlw7GzCA9IC81aPgjDAr+U6a15s2L4gd/FZ4Je2gnwKXS4Gm6K9wZWaJ5GDdkj6eKxTg9H+WO7nwyKQXlSmrjZW8lvvTbEwtvtYOLdhSL8g9/FgSqsDIbEZ2jGVy0eI+Jze5W2wg5HLXHCvgD8cbifTTpaXAdniXzcN5vlXGBH2KAT7Befe7y2Bd3k2iMZrvO6uTzcTw85xuOqmCekmWvC3rK7hQAYNzPfLYyLx9/4azlNx1htsxwwKbZKM01YPpffzUryxWTSLaaTfknxfruv5AJzmo5xZdjP9/1u7vHasQpv08ESx8nCBmjl8dwOh2MuuL+oedKlOc7U0g7RDHCKjrOCkF/5i7vC224a+KZZU/Lca+1IYhY/DZxkx8tsdpOwIjwhzTxh4q5KVa4K+lg5ySYX4F12PztIl23ipoi6Yts0YEh87vXaODkcZ51m8eLpTSHCjf6qOc8HdTgcyThhnycn2fHE7cY0yil2LNE5cLEZSTSVjVbsB+eYE+4hR0eCTa6PsprkTdwAisJuJZqmdIJmOUGzDIvHk9I05xmh3ZrjVDtGgPCYaWbY5bodjiLur2GeTIiJW/+nzkSF8RoNNn5OGsliSE8ZOG0RHqhitZ5Ei4aJdwEGyMXnmvpl8qF04IUIL0sDL5e1G80CVS4LD3CuRpcvC7zNDvJjs4LHvLm9N4ej3nDljvOkkv+2IlWlPBYDK8K3/R76SZFDyCKMYPiB18XBOfqt75JM4iogh/CgtJDDlAh/DuFus2xW9frVsE6znKsjpONO3BTR3dGf2H00OPtahwM4AlbsRpVX6BirNE+/pHhampa8lG0+TIjh+14XHwr7MWhxwPT/9VbWND2wX1J8PbWWTs3jo/TPs5FlXDzuNG28xQ4Vm47yCCN43OO18yCtXBIOslEnOIThLq+dnaTZaMfZI+nI2XEB2BRPpunmnAsAAAmYSURBVJqKJdrX2Cpu1e5w1FTYmzXk00EvzYSkUXIIY3j8vb+6OPD5aOBF08hfyfGs0wkMykvSMG1J4VIy3XDo2XKX18EeyXCBHaZZQ16WDL8zy8iKIYvhh/4qABrU8uFwL+/RLCFRHv5+aeWfvM5FbahZGscWh+PIp6bq885wH20ENMSOgg0obQRcHu6rZVhzIhThBdPIc6bpiBH1xWCbaeIxaaKTPGfqKP8u3M1VQV9JGuQ94QDH6wRplEaUFHC2jnLOLNwYK/GIaSFIMHAwwLOzGDbucNQzC6JAIvJ5EVERWTHzqw9zuo4lTLSJnj8WMKp0aY62hCHOS01aLeeFQ1wZ7OWS4EDFmE6041xuD9CA0hCL9gYd5wNhf/E4p+lYWW9rGuVCO//2ou2mgfuklRxCQJQOyiP8yKxY8Hy+w3G0Mu9UjIgcB1wM7Jh/OMcOp9lDvDfcVxxY3StpbvBW1aRFvVlDPhfsLs75zAMXBMN81+tmx5RyzTfaoTJzsRSwXrO0aUAIsc9leWKkYYE8cn7hd/KgtnKqHSOP8Lhpdq39DsckFmKJ83XgC8whxfmkNJXZUYVQMtS4HumKR8E1Y8mgpFDWapaPB32wRN7ek7kkPEBLvM8BkVBnUK4IB8riqbSSL5iLjeJVHOK9v4JHzVzYK2nu9tq512tzou5wTGFewi4ilwO7VXVrFa/9pIg8KCIPjsf52Fu8FQzhM4FggQmEITw2L9BQ5SOV88OhshYij2gu5xrNLXk8p+t44q1bOwEtU1bZz0ljom+kELkyIlLx58uZ/RQph8Mxe2Zc6ojInUCSN+mXgS8Bl1RzIlW9FrgWoNtkFCLXvq/FniSrNEe/pHlKmgiPonLHudBOmOjvaIFlhOxe4nhyCStsiMR46kbl3V4bZwajgC1+eXIIt5mOYhdpa4WUSysWiYdZOxyOxWNGYVfVi5KeF5FXAeuBrRL9oa4FHhaRc1S1r9oAQhEel2ZgaSa9Hwk8Kw2sj6tGJuOj7JzcqblE3GdaudgeLIknAF6QxrINyRHx+Ya/hjeHBzlJxxkWn7tNG9smTREaxGdFwrp9GM+JusOxBMw5OamqjwOrCo9FZDvwWlU9+moVl5gHzDLOtyO0EkxyhBS2SCujNcgX/860cZxmOVXHsUQr9UF8fuitTHz9sPhs9isXQP3KdHDFFLfHwqre4XAsPm7XqQZkxfBNfzVvCId4pR5iXDx+b5axtYI9wWJjRbjJ72Kl5lijOQbxeVkyc3ZdfNxrQYg82DsIGMLjdtPBQ17rwgbucDgSWTBhV9V1C3WsY4Fx8bjNX85tHDmDGwYkzcACpYIe81oiU64qh187HI6Fw63Y65RCo1BGLc+aRgYX0FpgVjhRdziWHCfsdcgGO86/DvcCUb5cLPxWlnHbIg/9cDgcRwauB7vO8FT5cLiXDEoGJR03QF2gw2yw47UOz+FwLAFO2OuMjTqeWJXuo5xlR5c8HofDsfQ4Ya8zvNgTfioGSDljW4fjmMAJe53xgjQmdrVmER41x04TmMNxLOOEvc7IiuEnppN8bGtriUT9WWnk6To3V3M4HBGuKqYOedRrZYdp4Ew7SqNanjJNvCANrvTQ4ThGcMJepxyQFHd6roXf4TgWcakYh8PhqDOcsDscDked4YTd4XA46gwn7A6Hw1FnOGF3OByOOsMJu8PhcNQZTtgdDoejznDC7nA4HHWGE3aHw+GoM5ywOxwOR53hhN3hcDjqDCfsDofDUWc4YXc4HI46wwm7w+Fw1BlO2B0Oh6POcMLucDgcdYYTdofD4agznLA7HA5HneGE3eFwOOqMeQu7iHxWRLaJyJMi8j8WIiiHw+FwzJ15DbMWkTcD7wTOUNWsiKxamLAcDofDMVfmu2L/FPDfVDULoKr98w/J4XA4HPNBVHXu/1nkUWAzcCkwAXxeVR+o8NpPAp+MH54CbJvziWdmBbBvEY+/2Lj4a8/R/h5c/LVlseI/QVVXzvSiGVMxInIn0J3woy/H/78DeB1wNvAjEdmgCVcLVb0WuHam8y0EIvKgqr52Kc61GLj4a8/R/h5c/LWl1vHPKOyqelGln4nIp4CfxkJ+v4hYoivVwMKF6HA4HI7ZMN8c+8+BtwCIyMlAmqP79snhcDiOeuZVFQNcD1wvIk8AOeAjSWmYGrAkKZ9FxMVfe4729+Diry01jX9em6cOh8PhOPJwnacOh8NRZzhhdzgcjjqjroW9HuwOROTzIqIisqLWscwGEfmqiDwjIo+JyM9EpL3WMVWDiFwaf2eeF5G/rHU8s0FEjhOR34jI0/F3/ppaxzQXRMQTkUdE5NZaxzIXRKRdRG6Ov/9Pi8h5Sx1D3Qr7FLuD04Gv1TikWSMixwEXAztqHcscuAN4paqeATwLfLHG8cyIiHjA3wNvB04D/lRETqttVLMiAP6Dqr6CqLfk00dZ/AWuAZ6udRDz4G+BX6nqqcCrqcF7qVthpz7sDr4OfAE46na4VfV2VQ3ih/cBa2sZT5WcAzyvqi+qag74f0SLg6MCVd2jqg/H/x4hEpQ1tY1qdojIWuBfAd+tdSxzQUSWAW8ArgNQ1ZyqHlzqOOpZ2E8GLhSRLSJyt4icXeuAZoOIXA7sVtWttY5lAbgK+GWtg6iCNcDOSY93cZQJYwERWQdsArbUNpJZ8w2ixYytdSBzZANRg+b34nTSd0WkeamDmG8de01ZKLuDWjFD/F8CLlnaiGbHdPGr6ub4NV8mShHctJSxzRFJeO6I+b5Ui4i0AD8B/kxVh2sdT7WIyGVAv6o+JCJvqnU8c8QHzgQ+q6pbRORvgb8EvrLUQRy1HO12B5XiF5FXAeuBrSICURrjYRE5R1X7ljDEaZnu9w8gIh8BLgPeeiRdUKdhF3DcpMdrgd4axTInRCRFJOo3qepPax3PLDkfuFxE3gE0AMtE5EZV/WCN45oNu4Bdqlq4U7qZSNiXlHpOxRy1dgeq+riqrlLVdaq6jujLcuaRJOozISKXAn8BXK6qY7WOp0oeAE4SkfUikgbeD9xS45iqRqJVwHXA06r6P2sdz2xR1S+q6tr4O/9+4F+OMlEn/hvdKSKnxE+9FXhqqeM4qlfsM3Ck2h0cK3wLyAB3xHcd96nq1bUNaXpUNRCRzwC3AR5wvao+WeOwZsP5wIeAx2NLbYAvqeo/1zCmY5HPAjfFi4MXgY8udQDOUsDhcDjqjHpOxTgcDscxiRN2h8PhqDOcsDscDked4YTd4XA46gwn7A6Hw1FnOGF3OByOOsMJu8PhcNQZ/x/3voZ3hKPVJAAAAABJRU5ErkJggg==\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "map_theta = ed.models.PointMass(tf.Variable(tf.zeros(2,dtype=tf.float64)))\n", "inference = ed.MAP({theta:map_theta}, {X:Xtr,W:np.ones(Xtr.shape[0],dtype=np.float64),y:ytr.reshape((ytr.shape[0]))})\n", "inference.run()\n", "mapx = np.linspace(-6,6,500)\n", "mapy = -(mapx*map_theta.eval()[0]) / map_theta.eval()[1]\n", "\n", "thetachain = ed.models.Empirical(params=tf.Variable(tf.zeros([n_mcsamples,nfeatures],dtype=tf.float64)))\n", "inference = ed.HMC({theta:thetachain},\n", " {X:Xtr,W:np.ones(Xtr.shape[0],dtype=np.float64),y:ytr.reshape((ytr.shape[0]))})\n", "inference.run(step_size=n_mcstepsize)\n", "thetahat = ed.models.Empirical(params = thetachain.params.eval()[n_mcburnin:n_mcsamples:n_mcthinning])\n", "y_post = ed.copy(y,{theta:thetahat})\n", "probs_BLR = [y_post.eval(feed_dict={X:xymeshgrid, W:np.ones(xymeshgrid.shape[0],dtype=np.float64)}) for _ in range(ppc_samples)]\n", "probs_BLR = np.array(probs_BLR).mean(axis=0).reshape(xmeshgrid.shape)\n", "\n", "plt.title('Bayes LR')\n", "plt.contourf(xmeshgrid, ymeshgrid, probs_BLR, 25, cmap=\"RdBu\")\n", "plt.scatter(Xtr[:,0],Xtr[:,1], c=ycolors)\n", "plt.plot(mapx,mapy)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Performing Bayesian Coreset\n", "\n", "We compute the GIGA coreset of the full dataset. This step includes evaluating an approximate posterior via a Laplace approximation, a discretization and random projection of the log-likelihood, the computation of the coreset and the selection of samples with non-zero weight (see previous notebook for an explanation of this code). At the end, we run MAP estimation and HMC on the coreset data and we visualize the result." ] }, { "cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 1s | Loss: 40.624\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 1s | Loss: 8.640\n", "10000/10000 [100%] ██████████████████████████████ Elapsed: 10s | Acceptance Rate: 1.000\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 11, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "qtheta = ed.models.MultivariateNormalTriL(\n", " loc = tf.Variable(tf.zeros(nfeatures,dtype=tf.float64)),\n", " scale_tril = tf.Variable(tf.eye(nfeatures,nfeatures,dtype=tf.float64)))\n", "inference = ed.Laplace({theta:qtheta}, {X:Xtr,W:np.ones(Xtr.shape[0],dtype=np.float64),y:ytr.reshape((ytr.shape[0]))})\n", "inference.run()\n", "\n", "t0 = time.time()\n", "randomprojection = ProjectionF(Xtr, grad_log_likelihood, nrandomdims, qtheta)\n", "vecs = randomprojection.get()\n", "bc_giga = bc.GIGA(vecs)\n", "bc_giga.run(ncoresamples)\n", "t_giga = time.time() - t0\n", "\n", "Wt = bc_giga.weights()\n", "Xwt = Xtr[Wt>0]\n", "ywt = ytr[Wt>0]\n", "wts = Wt[Wt>0]\n", "\n", "map_theta = ed.models.PointMass(tf.Variable(tf.zeros(2,dtype=tf.float64)))\n", "inference = ed.MAP({theta:map_theta}, {X:Xwt,W:wts/10.0,y:ywt.reshape((ywt.shape[0]))})\n", "inference.run()\n", "mapx = np.linspace(-6,6,500)\n", "mapy = -(mapx*map_theta.eval()[0]) / map_theta.eval()[1]\n", "\n", "thetachain = ed.models.Empirical(params=tf.Variable(tf.zeros([n_mcsamples,nfeatures],dtype=tf.float64)))\n", "inference = ed.HMC({theta:thetachain},\n", " {X:Xwt,W:wts/10.0,y:ywt.reshape((ywt.shape[0]))})\n", "inference.run(step_size=n_mcstepsize)\n", "thetahat = ed.models.Empirical(params = thetachain.params.eval()[n_mcburnin:n_mcsamples:n_mcthinning])\n", "y_post = ed.copy(y,{theta:thetahat}) \n", "probs_BLR = [y_post.eval(feed_dict={X:xymeshgrid, W:np.ones(xymeshgrid.shape[0],dtype=np.float64)}) for _ in range(ppc_samples)]\n", "probs_BLR = np.array(probs_BLR).mean(axis=0).reshape(xmeshgrid.shape)\n", "\n", "plt.title('GIGA coreset')\n", "ycolors_giga = np.array(['r']*ywt.shape[0])\n", "ycolors_giga[ywt[:,0]==1] = 'b'\n", "plt.contourf(xmeshgrid, ymeshgrid, probs_BLR, 25, cmap=\"RdBu\")\n", "plt.scatter(Xwt[:,0],Xwt[:,1], c=ycolors_giga, s=wts)\n", "plt.plot(mapx,mapy)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Discussion\n", "All the models (standard logistic regression, Bayesian logistic regression, and Bayesian logistic regression on the coreset) running on the whole dataset are able to identify a good discriminator." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3. Parallel Computation of Coresets: Random Subsets\n", "\n", "In this third part, we randomly partition the data in four subsets. We can safely assume that each one of these datasets come from the same distribution. We compute the four coresets in parallel, then we aggregate them and examine the result." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Partitioning the data\n", "\n", "We divide the training data in four training sub-datasets." ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [], "source": [ "partitioning = np.random.permutation(Xtr.shape[0])\n", "\n", "Xtr1 = Xtr[partitioning[0:150]]\n", "ytr1 = ytr[partitioning[0:150]]\n", "\n", "Xtr2 = Xtr[partitioning[150:300]]\n", "ytr2 = ytr[partitioning[150:300]]\n", "\n", "Xtr3 = Xtr[partitioning[300:450]]\n", "ytr3 = ytr[partitioning[300:450]]\n", "\n", "Xtr4 = Xtr[partitioning[450:600]]\n", "ytr4 = ytr[partitioning[450:600]]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Computing the coresets\n", "We compute the four coresets in parallel. We implement a helper function (*compute_parallel_coreset()*) that computes the coreset for a given sub-dataset. Notice that each of the four coresets selects 10 points, so that, overall, we have the same number of points as when we computed the coreset on the whole dataset (40). At the end, we collect the results and plot them." ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 1s | Loss: 13.498\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 3.452\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 9.804\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 1s | Loss: 6.252\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 16.456\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 3.158\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 17.184\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 3s | Loss: 7.039\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 13, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "def compute_parallel_coreset(Xtr,ytr):\n", " qtheta = ed.models.MultivariateNormalTriL(\n", " loc = tf.Variable(tf.zeros(nfeatures,dtype=tf.float64)),\n", " scale_tril = tf.Variable(tf.eye(nfeatures,nfeatures,dtype=tf.float64)))\n", " inference = ed.Laplace({theta:qtheta}, {X:Xtr,W:np.ones(Xtr.shape[0],dtype=np.float64),y:ytr.reshape((ytr.shape[0]))})\n", " inference.run()\n", " \n", " t0 = time.time()\n", " randomprojection = ProjectionF(Xtr, grad_log_likelihood, nrandomdims, qtheta)\n", " vecs = randomprojection.get()\n", " bc_giga = bc.GIGA(vecs)\n", " bc_giga.run(ncoresamples)\n", " t_giga = time.time() - t0\n", "\n", " Wt = bc_giga.weights()\n", " Xwt = Xtr[Wt>0]\n", " ywt = ytr[Wt>0]\n", " wts = Wt[Wt>0]\n", " ycolors = np.array(['r']*ywt.shape[0])\n", " ycolors[ywt[:,0]==1] = 'b'\n", " \n", " map_theta = ed.models.PointMass(tf.Variable(tf.zeros(2,dtype=tf.float64)))\n", " inference = ed.MAP({theta:map_theta}, {X:Xwt,W:wts/10.0,y:ywt.reshape((ywt.shape[0]))})\n", " inference.run()\n", " mapx = np.linspace(-6,6,500)\n", " mapy = -(mapx*map_theta.eval()[0]) / map_theta.eval()[1]\n", " \n", " return Xwt,ywt,wts,mapx,mapy,ycolors,t_giga\n", "\n", "ncoresamples = 10\n", "Xwt1,ywt1,wts1,mapx1,mapy1,ycolors1,t1 = compute_parallel_coreset(Xtr1,ytr1)\n", "Xwt2,ywt2,wts2,mapx2,mapy2,ycolors2,t2 = compute_parallel_coreset(Xtr2,ytr2)\n", "Xwt3,ywt3,wts3,mapx3,mapy3,ycolors3,t3 = compute_parallel_coreset(Xtr3,ytr3)\n", "Xwt4,ywt4,wts4,mapx4,mapy4,ycolors4,t4 = compute_parallel_coreset(Xtr4,ytr4)\n", "\n", "fig, axes = plt.subplots(2,2)\n", "fig.suptitle('Parallel coresets')\n", "axes[0,0].scatter(Xwt1[:,0],Xwt1[:,1], c=ycolors1, s=wts1)\n", "axes[0,0].plot(mapx1,mapy1)\n", "axes[0,0].axhline(y=0, color='gray')\n", "axes[0,0].axvline(x=0, color='gray')\n", "\n", "axes[0,1].scatter(Xwt2[:,0],Xwt2[:,1], c=ycolors2, s=wts2)\n", "axes[0,1].plot(mapx2,mapy2)\n", "axes[0,1].axhline(y=0, color='gray')\n", "axes[0,1].axvline(x=0, color='gray')\n", "\n", "axes[1,0].scatter(Xwt3[:,0],Xwt3[:,1], c=ycolors3, s=wts3)\n", "axes[1,0].plot(mapx3,mapy3)\n", "axes[1,0].axhline(y=0, color='gray')\n", "axes[1,0].axvline(x=0, color='gray')\n", "\n", "axes[1,1].scatter(Xwt4[:,0],Xwt4[:,1], c=ycolors4, s=wts4)\n", "axes[1,1].plot(mapx4,mapy4)\n", "axes[1,1].axhline(y=0, color='gray')\n", "axes[1,1].axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Assembling the data\n", "We now collect all the coresets in a single dataset, and we perform Bayesian analysis as we did before." ] }, { "cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 2s | Loss: 11.286\n", "10000/10000 [100%] ██████████████████████████████ Elapsed: 11s | Acceptance Rate: 1.000\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 14, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "Xpar = np.vstack((Xwt1,Xwt2,Xwt3,Xwt4))\n", "ypar = np.vstack((ywt1,ywt2,ywt3,ywt4))\n", "wpar = np.vstack((wts1.reshape((wts1.shape[0],1)),wts2.reshape((wts2.shape[0],1)),wts3.reshape((wts3.shape[0],1)),wts4.reshape((wts4.shape[0],1))))\n", "wpar = wpar.reshape((wpar.shape[0]))\n", "\n", "map_theta = ed.models.PointMass(tf.Variable(tf.zeros(2,dtype=tf.float64)))\n", "inference = ed.MAP({theta:map_theta}, {X:Xpar,W:wpar/10.0,y:ypar.reshape((ypar.shape[0]))})\n", "inference.run()\n", "mapx = np.linspace(-6,6,500)\n", "mapy = -(mapx*map_theta.eval()[0]) / map_theta.eval()[1]\n", "\n", "thetachain = ed.models.Empirical(params=tf.Variable(tf.zeros([n_mcsamples,nfeatures],dtype=tf.float64)))\n", "inference = ed.HMC({theta:thetachain},\n", " {X:Xpar,W:wpar,y:ypar.reshape((ypar.shape[0]))})\n", "inference.run(step_size=n_mcstepsize)\n", "thetahat = ed.models.Empirical(params = thetachain.params.eval()[n_mcburnin:n_mcsamples:n_mcthinning])\n", "y_post = ed.copy(y,{theta:thetahat})\n", " \n", "probs_BLR = [y_post.eval(feed_dict={X:xymeshgrid, W:np.ones(xymeshgrid.shape[0],dtype=np.float64)}) for _ in range(ppc_samples)]\n", "probs_BLR = np.array(probs_BLR).mean(axis=0).reshape(xmeshgrid.shape)\n", "\n", "plt.figure()\n", "plt.title('GIGA coreset (aggregated dataset)')\n", "ycolors_giga = np.array(['r']*ypar.shape[0])\n", "ycolors_giga[ypar[:,0]==1] = 'b'\n", "plt.contourf(xmeshgrid, ymeshgrid, probs_BLR, 25, cmap=\"RdBu\")\n", "plt.scatter(Xpar[:,0],Xpar[:,1], c=ycolors_giga, s=wpar)\n", "plt.plot(mapx,mapy)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Printing out coreset training times\n", "We compare the time to train a single coreset on the whole dataset and training four coresets in parallel." ] }, { "cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Time to compute a single coreset on 600 samples: 22.029346466064453 secs\n", "Total time to compute four coresets on 150 samples: 226.08742094039917 secs\n", "Individual coreset times: [29.29443860054016, 44.80942702293396, 65.24506831169128, 86.73848700523376]\n", "Average time for the four coresets: 56.52185523509979 secs\n" ] } ], "source": [ "print('Time to compute a single coreset on {0} samples: {1} secs'.format(Xtr.shape[0],t_giga))\n", "print('Total time to compute four coresets on {0} samples: {1} secs'.format(150,t1+t2+t3+t4))\n", "print('Individual coreset times: [{0}, {1}, {2}, {3}]'.format(t1,t2,t3,t4))\n", "print('Average time for the four coresets: {0} secs'.format((t1+t2+t3+t4)/4.0))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Discussion\n", "On a simple dataset, random subsampling produces training sub-datasets that still represent the original distribution quite closely. The MAP discriminator associated with the sub-datasets shows a certain degree of variance, with some instances closer to the baseline (third sub-dataset) and other further from the baseline (second sub-dataset). However, sssembling the four coresets returns a posterior discriminator very close to the one computed from the whole dataset.\n", "Computing time seems not to be sensibly affected by the partitioning of the data. (?) " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 4. Parallel Computation of Coresets: Non i.d. Subsets\n", "In this fourth part, we consider a more challenging partition of the data, in which the data are divided in two sub-datasets according to their generating clusters. This induces two different distributions. We compute the two coresets in parallel, then we aggregate them and examine the result." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Partitioning the data\n", "\n", "We divide the training data in two training sub-datasets. Additionally, we plot the data to show that the two datasets are generated by different distributions. " ] }, { "cell_type": "code", "execution_count": 16, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "" ] }, "execution_count": 16, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "Xtr12 = np.vstack((X_pdf1,X_pdf2))\n", "ytr12 = np.vstack((y_pdf1,y_pdf2))\n", "permutation = np.random.permutation(nsamples*2)\n", "Xtr12 = Xtr12[permutation]\n", "ytr12 = ytr12[permutation]\n", " \n", "Xtr13 = np.vstack((X_pdf1,X_pdf3))\n", "ytr13 = np.vstack((y_pdf1,y_pdf3))\n", "permutation = np.random.permutation(nsamples*2)\n", "Xtr13 = Xtr13[permutation]\n", "ytr13 = ytr13[permutation]\n", "\n", "fig, axes = plt.subplots(1,2)\n", "ycolors_12 = np.array(['r']*ytr12.shape[0])\n", "ycolors_12[ytr12[:,0]==1] = 'b'\n", "axes[0].set_title('First sub-dataset')\n", "axes[0].scatter(Xtr12[:,0],Xtr12[:,1], c=ycolors_12)\n", "axes[0].axhline(y=0, color='gray')\n", "axes[0].axvline(x=0, color='gray')\n", "\n", "ycolors_13 = np.array(['r']*ytr13.shape[0])\n", "ycolors_13[ytr13[:,0]==1] = 'b'\n", "axes[1].set_title('First sub-dataset')\n", "axes[1].scatter(Xtr13[:,0],Xtr13[:,1], c=ycolors_13)\n", "axes[1].axhline(y=0, color='gray')\n", "axes[1].axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Computing the coresets\n", "We compute the two coresets in parallel. We rely on the previous helper function (*compute_parallel_coreset()*). Notice that each of the two coreset selects 20 points, so that, overall, we have the same number of points as when we computed the coreset on the whole dataset (40). At the end, we collect the results and plot them." ] }, { "cell_type": "code", "execution_count": 17, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 3s | Loss: 26.856\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 3s | Loss: 3.843\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 4s | Loss: 26.399\n", "1000/1000 [100%] ██████████████████████████████ Elapsed: 4s | Loss: 4.593\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 17, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "ncoresamples = 20\n", "Xwt12,ywt12,wts12,mapx12,mapy12,ycolors12,t12 = compute_parallel_coreset(Xtr12,ytr12)\n", "Xwt13,ywt13,wts13,mapx13,mapy13,ycolors13,t13 = compute_parallel_coreset(Xtr13,ytr13)\n", "\n", "fig, axes = plt.subplots(1,2)\n", "fig.suptitle('Parallel coresets')\n", "axes[0].scatter(Xwt12[:,0],Xwt12[:,1], c=ycolors12, s=wts12)\n", "axes[0].plot(mapx12,mapy12)\n", "axes[0].axhline(y=0, color='gray')\n", "axes[0].axvline(x=0, color='gray')\n", "\n", "axes[1].scatter(Xwt13[:,0],Xwt13[:,1], c=ycolors13, s=wts13)\n", "axes[1].plot(mapx13,mapy13)\n", "axes[1].axhline(y=0, color='gray')\n", "axes[1].axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Assembling the data\n", "We now collect all the coresets in a single dataset, and we perform Bayesian analysis as we did before." ] }, { "cell_type": "code", "execution_count": 19, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "/home/fmzennaro/miniconda2_1/envs/bayes3/lib/python3.6/site-packages/edward/util/random_variables.py:52: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n", " not np.issubdtype(value.dtype, np.float) and \\\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "1000/1000 [100%] ██████████████████████████████ Elapsed: 4s | Loss: 8.286\n", "10000/10000 [100%] ██████████████████████████████ Elapsed: 17s | Acceptance Rate: 1.000\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 19, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "Xparr = np.vstack((Xwt12,Xwt13))\n", "yparr = np.vstack((ywt12,ywt13))\n", "wparr = np.vstack((wts12.reshape((wts12.shape[0],1)),wts13.reshape((wts13.shape[0],1))))\n", "wparr = wparr.reshape((wparr.shape[0]))\n", "\n", "map_theta = ed.models.PointMass(tf.Variable(tf.zeros(2,dtype=tf.float64)))\n", "inference = ed.MAP({theta:map_theta}, {X:Xparr,W:wparr/10.0,y:yparr.reshape((yparr.shape[0]))})\n", "inference.run()\n", "mapx = np.linspace(-6,6,500)\n", "mapy = -(mapx*map_theta.eval()[0]) / map_theta.eval()[1]\n", "\n", "thetachain = ed.models.Empirical(params=tf.Variable(tf.zeros([n_mcsamples,nfeatures],dtype=tf.float64)))\n", "inference = ed.HMC({theta:thetachain},\n", " {X:Xparr,W:wparr/10.0,y:yparr.reshape((yparr.shape[0]))})\n", "inference.run(step_size=n_mcstepsize)\n", "thetahat = ed.models.Empirical(params = thetachain.params.eval()[n_mcburnin:n_mcsamples:n_mcthinning])\n", "y_post = ed.copy(y,{theta:thetahat})\n", " \n", "probs_BLR = [y_post.eval(feed_dict={X:xymeshgrid, W:np.ones(xymeshgrid.shape[0],dtype=np.float64)}) for _ in range(ppc_samples)]\n", "probs_BLR = np.array(probs_BLR).mean(axis=0).reshape(xmeshgrid.shape)\n", "\n", "plt.figure()\n", "plt.title('GIGA coreset (aggregated dataset)')\n", "ycolors_giga = np.array(['r']*yparr.shape[0])\n", "ycolors_giga[yparr[:,0]==1] = 'b'\n", "plt.contourf(xmeshgrid, ymeshgrid, probs_BLR, 25, cmap=\"RdBu\")\n", "plt.scatter(Xparr[:,0],Xparr[:,1], c=ycolors_giga, s=wparr)\n", "plt.plot(mapx,mapy)\n", "plt.axhline(y=0, color='gray')\n", "plt.axvline(x=0, color='gray')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Printing out coreset training times\n", "We compare the time to train a single coreset on the whole dataset and training four coresets in parallel." ] }, { "cell_type": "code", "execution_count": 20, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Time to compute a single coreset on 600 samples: 22.029346466064453 secs\n", "Total time to compute two coresets on 150 samples: 267.29918479919434 secs\n", "Individual coreset times: [120.1095643043518, 147.18962049484253]\n", "Average time for the four coresets: 133.64959239959717 secs\n" ] } ], "source": [ "print('Time to compute a single coreset on {0} samples: {1} secs'.format(Xtr.shape[0],t_giga))\n", "print('Total time to compute two coresets on {0} samples: {1} secs'.format(150,t12+t13))\n", "print('Individual coreset times: [{0}, {1}]'.format(t12,t13))\n", "print('Average time for the four coresets: {0} secs'.format((t12+t13)/2.0))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Discussion\n", "When partitioning the data according to their generating clusters, we define two sub-datasets with different distribution. Not surprisingly, the MAP discriminators associated with the two sub-datasets are sensibly different. However, once again, assembling the coresets returns a posterior discriminator very close to the one computed from the whole dataset.\n", "Computing time seems not to be sensibly affected by the partitioning of the data. (?) " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## References\n", "\n", "\\[1\\] Campbell, T. & Broderick, T. Automated Scalable Bayesian Inference via Hilbert Coresets arXiv preprint arXiv:1710.05053, 2017.\n", "\n", "\\[2\\] Campbell, T. & Broderick, T. Bayesian coreset construction via greedy iterative geodesic ascent arXiv preprint arXiv:1802.01737, 2018\n", "\n", "\\[3\\] [Bayesian Coresets: Automated, Scalable Inference](https://github.com/trevorcampbell/bayesian-coresets)\n", "\n", "\\[4\\] Tran, D.; Kucukelbir, A.; Dieng, A. B.; Rudolph, M.; Liang, D. & Blei, D. M. Edward: A library for probabilistic modeling, inference, and criticism arXiv preprint arXiv:1610.09787, 2016" ] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.7.7" } }, "nbformat": 4, "nbformat_minor": 2 }