|
| 1 | +{ |
| 2 | + "cells": [ |
| 3 | + { |
| 4 | + "cell_type": "markdown", |
| 5 | + "id": "2c795b88-378b-489d-85bb-3e1786930b4a", |
| 6 | + "metadata": {}, |
| 7 | + "source": [ |
| 8 | + "# Test sampling algorithms" |
| 9 | + ] |
| 10 | + }, |
| 11 | + { |
| 12 | + "cell_type": "code", |
| 13 | + "execution_count": null, |
| 14 | + "id": "eafc9aac-719c-434f-8816-7aae59f171db", |
| 15 | + "metadata": { |
| 16 | + "execution": { |
| 17 | + "iopub.execute_input": "2025-05-08T16:21:49.363904Z", |
| 18 | + "iopub.status.busy": "2025-05-08T16:21:49.363456Z", |
| 19 | + "iopub.status.idle": "2025-05-08T16:21:50.671986Z", |
| 20 | + "shell.execute_reply": "2025-05-08T16:21:50.671739Z" |
| 21 | + } |
| 22 | + }, |
| 23 | + "outputs": [], |
| 24 | + "source": [ |
| 25 | + "import matplotlib.pyplot as plt\n", |
| 26 | + "import numpy as np\n", |
| 27 | + "\n", |
| 28 | + "import ment" |
| 29 | + ] |
| 30 | + }, |
| 31 | + { |
| 32 | + "cell_type": "code", |
| 33 | + "execution_count": null, |
| 34 | + "id": "b2fda8d3", |
| 35 | + "metadata": {}, |
| 36 | + "outputs": [], |
| 37 | + "source": [ |
| 38 | + "plt.style.use(\"../style.mplstyle\")" |
| 39 | + ] |
| 40 | + }, |
| 41 | + { |
| 42 | + "cell_type": "markdown", |
| 43 | + "id": "e5922304-a9c9-47a7-b09f-7ae4f4845d2d", |
| 44 | + "metadata": {}, |
| 45 | + "source": [ |
| 46 | + "## Create distribution" |
| 47 | + ] |
| 48 | + }, |
| 49 | + { |
| 50 | + "cell_type": "code", |
| 51 | + "execution_count": null, |
| 52 | + "id": "3b8a7941-17c2-446d-aa36-6a3cb7308de2", |
| 53 | + "metadata": { |
| 54 | + "execution": { |
| 55 | + "iopub.execute_input": "2025-05-08T16:21:50.676745Z", |
| 56 | + "iopub.status.busy": "2025-05-08T16:21:50.676655Z", |
| 57 | + "iopub.status.idle": "2025-05-08T16:21:52.023604Z", |
| 58 | + "shell.execute_reply": "2025-05-08T16:21:52.023261Z" |
| 59 | + } |
| 60 | + }, |
| 61 | + "outputs": [], |
| 62 | + "source": [ |
| 63 | + "class RingDistribution:\n", |
| 64 | + " def __init__(self) -> None:\n", |
| 65 | + " self.ndim = 2\n", |
| 66 | + "\n", |
| 67 | + " def prob(self, x: np.ndarray) -> np.ndarray:\n", |
| 68 | + " x1 = x[:, 0]\n", |
| 69 | + " x2 = x[:, 1]\n", |
| 70 | + " log_prob = np.sin(np.pi * x1) - 2.0 * (x1**2 + x2**2 - 2.0) ** 2\n", |
| 71 | + " return np.exp(log_prob)\n", |
| 72 | + "\n", |
| 73 | + " def prob_grid(\n", |
| 74 | + " self, shape: tuple[int], limits: list[tuple[float, float]]\n", |
| 75 | + " ) -> tuple[np.ndarray, list[np.ndarray]]:\n", |
| 76 | + " edges = [\n", |
| 77 | + " np.linspace(limits[i][0], limits[i][1], shape[i] + 1)\n", |
| 78 | + " for i in range(self.ndim)\n", |
| 79 | + " ]\n", |
| 80 | + " coords = [0.5 * (e[:-1] + e[1:]) for e in edges]\n", |
| 81 | + " points = np.stack(\n", |
| 82 | + " [c.ravel() for c in np.meshgrid(*coords, indexing=\"ij\")], axis=-1\n", |
| 83 | + " )\n", |
| 84 | + " values = self.prob(points)\n", |
| 85 | + " values = values.reshape(shape)\n", |
| 86 | + " return values, coords" |
| 87 | + ] |
| 88 | + }, |
| 89 | + { |
| 90 | + "cell_type": "code", |
| 91 | + "execution_count": null, |
| 92 | + "id": "fe88acb4-1a21-4565-81f2-e41de80586bc", |
| 93 | + "metadata": {}, |
| 94 | + "outputs": [], |
| 95 | + "source": [ |
| 96 | + "ndim = 2\n", |
| 97 | + "xmax = 3.0\n", |
| 98 | + "dist = RingDistribution()" |
| 99 | + ] |
| 100 | + }, |
| 101 | + { |
| 102 | + "cell_type": "code", |
| 103 | + "execution_count": null, |
| 104 | + "id": "e083ac94", |
| 105 | + "metadata": {}, |
| 106 | + "outputs": [], |
| 107 | + "source": [ |
| 108 | + "grid_limits = 2 * [(-xmax, xmax)]\n", |
| 109 | + "grid_shape = (128, 128)\n", |
| 110 | + "grid_values, grid_coords = dist.prob_grid(grid_shape, grid_limits)\n", |
| 111 | + "\n", |
| 112 | + "fig, ax = plt.subplots(figsize=(2.5, 2.5))\n", |
| 113 | + "ax.pcolormesh(grid_coords[0], grid_coords[1], grid_values.T)\n", |
| 114 | + "plt.show()" |
| 115 | + ] |
| 116 | + }, |
| 117 | + { |
| 118 | + "cell_type": "code", |
| 119 | + "execution_count": null, |
| 120 | + "id": "4a72fdfd", |
| 121 | + "metadata": {}, |
| 122 | + "outputs": [], |
| 123 | + "source": [ |
| 124 | + "def plot_samples(x_pred: np.ndarray) -> tuple:\n", |
| 125 | + " fig, axs = plt.subplots(ncols=2, figsize=(5.0, 2.75), sharex=True, sharey=True)\n", |
| 126 | + " hist, edges = np.histogramdd(x_pred, bins=80, range=grid_limits)\n", |
| 127 | + " axs[0].pcolormesh(edges[0], edges[1], hist.T)\n", |
| 128 | + " axs[1].pcolormesh(grid_coords[0], grid_coords[1], grid_values.T)\n", |
| 129 | + " axs[0].set_title(\"PRED\", fontsize=\"medium\")\n", |
| 130 | + " axs[1].set_title(\"TRUE\", fontsize=\"medium\")\n", |
| 131 | + " return fig, axs" |
| 132 | + ] |
| 133 | + }, |
| 134 | + { |
| 135 | + "cell_type": "code", |
| 136 | + "execution_count": null, |
| 137 | + "id": "f5ff2f9d", |
| 138 | + "metadata": {}, |
| 139 | + "outputs": [], |
| 140 | + "source": [ |
| 141 | + "def evaluate_sampler(sampler, size: int = 100_000):\n", |
| 142 | + " x_pred = sampler(dist.prob, size=size)\n", |
| 143 | + " return plot_samples(x_pred)" |
| 144 | + ] |
| 145 | + }, |
| 146 | + { |
| 147 | + "cell_type": "markdown", |
| 148 | + "id": "18c50faa", |
| 149 | + "metadata": {}, |
| 150 | + "source": [ |
| 151 | + "## Grid sampler" |
| 152 | + ] |
| 153 | + }, |
| 154 | + { |
| 155 | + "cell_type": "code", |
| 156 | + "execution_count": null, |
| 157 | + "id": "09414cdc", |
| 158 | + "metadata": {}, |
| 159 | + "outputs": [], |
| 160 | + "source": [ |
| 161 | + "sampler = ment.GridSampler(grid_limits=grid_limits, grid_shape=grid_shape)\n", |
| 162 | + "\n", |
| 163 | + "evaluate_sampler(sampler);" |
| 164 | + ] |
| 165 | + }, |
| 166 | + { |
| 167 | + "cell_type": "markdown", |
| 168 | + "id": "1cfb7494", |
| 169 | + "metadata": {}, |
| 170 | + "source": [ |
| 171 | + "## Metropolis-Hastings sampler" |
| 172 | + ] |
| 173 | + }, |
| 174 | + { |
| 175 | + "cell_type": "code", |
| 176 | + "execution_count": null, |
| 177 | + "id": "37e61a8a", |
| 178 | + "metadata": {}, |
| 179 | + "outputs": [], |
| 180 | + "source": [ |
| 181 | + "chains = 4\n", |
| 182 | + "proposal_cov = np.identity(ndim) * 0.25\n", |
| 183 | + "start_loc = np.zeros(ndim)\n", |
| 184 | + "start_cov = np.identity(ndim) * 0.25\n", |
| 185 | + "start_point = np.random.multivariate_normal(start_loc, start_cov, size=chains)\n", |
| 186 | + "\n", |
| 187 | + "sampler = ment.samp.MetropolisHastingsSampler(\n", |
| 188 | + " ndim=ndim,\n", |
| 189 | + " proposal_cov=proposal_cov,\n", |
| 190 | + " start=start_point,\n", |
| 191 | + " chains=chains,\n", |
| 192 | + " burnin=0,\n", |
| 193 | + ")\n", |
| 194 | + "evaluate_sampler(sampler);\n" |
| 195 | + ] |
| 196 | + }, |
| 197 | + { |
| 198 | + "cell_type": "code", |
| 199 | + "execution_count": null, |
| 200 | + "id": "00ab115a", |
| 201 | + "metadata": {}, |
| 202 | + "outputs": [], |
| 203 | + "source": [] |
| 204 | + } |
| 205 | + ], |
| 206 | + "metadata": { |
| 207 | + "kernelspec": { |
| 208 | + "display_name": "ment", |
| 209 | + "language": "python", |
| 210 | + "name": "python3" |
| 211 | + }, |
| 212 | + "language_info": { |
| 213 | + "codemirror_mode": { |
| 214 | + "name": "ipython", |
| 215 | + "version": 3 |
| 216 | + }, |
| 217 | + "file_extension": ".py", |
| 218 | + "mimetype": "text/x-python", |
| 219 | + "name": "python", |
| 220 | + "nbconvert_exporter": "python", |
| 221 | + "pygments_lexer": "ipython3", |
| 222 | + "version": "3.13.9" |
| 223 | + } |
| 224 | + }, |
| 225 | + "nbformat": 4, |
| 226 | + "nbformat_minor": 5 |
| 227 | +} |
0 commit comments