Skip to content

Commit fbf076d

Browse files
jacobdenobeljacobdenobel
committed
paper scripts
1 parent 3ab9ad5 commit fbf076d

File tree

7 files changed

+714
-2074
lines changed

7 files changed

+714
-2074
lines changed

‎scripts/distributions/cma.ipynb‎

Lines changed: 0 additions & 237 deletions
This file was deleted.

‎scripts/distributions/ecdf_plots.ipynb‎

Lines changed: 434 additions & 0 deletions
Large diffs are not rendered by default.

‎scripts/distributions/es.ipynb‎

Lines changed: 0 additions & 1233 deletions
This file was deleted.

‎scripts/distributions/opo.ipynb‎

Lines changed: 0 additions & 179 deletions
This file was deleted.

‎scripts/distributions/overview.ipynb‎

Lines changed: 0 additions & 397 deletions
This file was deleted.

scripts/distributions/samplers.ipynb renamed to scripts/distributions/plots.ipynb

Lines changed: 276 additions & 22 deletions
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,10 @@
1010
"import pandas as pd\n",
1111
"import matplotlib.pyplot as plt\n",
1212
"import seaborn as sns \n",
13-
"from modcma.c_maes import sampling\n",
13+
"from modcma.c_maes import (\n",
14+
" sampling\n",
15+
" ModularCMAES, parameters, options, constants, utils, es\n",
16+
")\n",
1417
"\n",
1518
"\n",
1619
"import matplotlib\n",
@@ -93,24 +96,6 @@
9396
"plt.savefig(\"figures/distributions_z.pdf\")\n"
9497
]
9598
},
96-
{
97-
"cell_type": "code",
98-
"execution_count": 7,
99-
"metadata": {},
100-
"outputs": [
101-
{
102-
"data": {
103-
"text/plain": [
104-
"(0.12156862745098039, 0.4666666666666667, 0.7058823529411765)"
105-
]
106-
},
107-
"execution_count": 7,
108-
"metadata": {},
109-
"output_type": "execute_result"
110-
}
111-
],
112-
"source": []
113-
},
11499
{
115100
"cell_type": "code",
116101
"execution_count": 4,
@@ -228,13 +213,282 @@
228213
},
229214
{
230215
"cell_type": "code",
231-
"execution_count": 6,
216+
"execution_count": null,
217+
"metadata": {},
218+
"outputs": [],
219+
"source": [
220+
"d = 2 \n",
221+
"\n",
222+
"def sphere(x):\n",
223+
" x = np.asarray(x)\n",
224+
" return x.dot(x)\n",
225+
"\n",
226+
"def get_meshgrid(objective_function, lb, ub, delta: float = 0.025):\n",
227+
" x = np.arange(lb, ub + delta, delta)\n",
228+
" y = np.arange(lb, ub + delta, delta)\n",
229+
"\n",
230+
" if hasattr(objective_function, \"optimum\"):\n",
231+
" xo, yo = objective_function.optimum.x\n",
232+
" x = np.sort(np.r_[x, xo])\n",
233+
" y = np.sort(np.r_[y, yo])\n",
234+
"\n",
235+
" X, Y = np.meshgrid(x, y)\n",
236+
"\n",
237+
" Z = np.zeros(X.shape)\n",
238+
" for idx1 in range(X.shape[0]):\n",
239+
" for idx2 in range(X.shape[1]):\n",
240+
" Z[idx1, idx2] = objective_function([X[idx1, idx2], Y[idx1, idx2]])\n",
241+
" return X, Y, Z\n",
242+
"\n",
243+
"\n",
244+
"X, Y, Z = get_meshgrid(sphere, -5, 3)\n",
245+
"\n",
246+
"x0 = np.array([-4, -4])\n",
247+
"\n",
248+
"\n",
249+
"modules = parameters.Modules()\n",
250+
"modules.sample_transformation = options.SampleTranformerType(1)\n",
251+
"settings = parameters.Settings(dim=2, modules=modules, x0=x0, sigma0=2)\n",
252+
"\n",
253+
"utils.set_seed(10)\n",
254+
"cma = ModularCMAES(settings)\n",
255+
"\n",
256+
"\n",
257+
"f, (ax1, ax2) = plt.subplots(1, 2, figsize=(8, 5), sharex=True, sharey=True)\n",
258+
"\n",
259+
"ax1.contourf(\n",
260+
" X, Y, np.log10(Z), levels=200, cmap=\"Spectral\", zorder=-1, vmin=-1, vmax=2.5\n",
261+
")\n",
262+
"ax2.contourf(\n",
263+
" X, Y, np.log10(Z), levels=200, cmap=\"Spectral\", zorder=-1, vmin=-1, vmax=2.5\n",
264+
")\n",
265+
"\n",
266+
"for i in range(3):\n",
267+
" m = cma.p.adaptation.m.copy()\n",
268+
" C = cma.p.adaptation.C.copy()\n",
269+
" sigma = cma.p.mutation.sigma\n",
270+
" theta = np.degrees(np.arctan2(C[1, 0], C[0, 0]))\n",
271+
" \n",
272+
" color = 'black'\n",
273+
" p = ax1.scatter(*m, label=i, color=color)\n",
274+
" \n",
275+
" current = Ellipse(\n",
276+
" m,\n",
277+
" *(sigma * np.diag(C)),\n",
278+
" angle=theta,\n",
279+
" facecolor=\"none\",\n",
280+
" edgecolor=p.get_edgecolor(),\n",
281+
" linewidth=2,\n",
282+
" linestyle=\"dashed\",\n",
283+
" zorder=0,\n",
284+
" )\n",
285+
" ax1.add_patch(current)\n",
286+
" cma.step(sphere)\n",
287+
" \n",
288+
" \n",
289+
" \n",
290+
"modules = parameters.Modules()\n",
291+
"modules.sample_transformation = options.SampleTranformerType(2)\n",
292+
"settings = parameters.Settings(dim=2, modules=modules, x0=x0, sigma0=2)\n",
293+
"\n",
294+
"utils.set_seed(10)\n",
295+
"cma = ModularCMAES(settings)\n",
296+
"\n",
297+
"for i in range(3):\n",
298+
" m = cma.p.adaptation.m.copy()\n",
299+
" C = cma.p.adaptation.C.copy()\n",
300+
" sigma = cma.p.mutation.sigma\n",
301+
" theta = np.degrees(np.arctan2(C[1, 0], C[0, 0]))\n",
302+
" \n",
303+
" color = 'black'\n",
304+
" p = ax2.scatter(*m, label=i, color=color)\n",
305+
" \n",
306+
" \n",
307+
" width = sigma * C[0, 0]\n",
308+
" height = sigma * C[1, 1]\n",
309+
" \n",
310+
" current = Rectangle(\n",
311+
" (-width / 2, -height / 2), width, height,\n",
312+
" facecolor=\"none\",\n",
313+
" edgecolor=p.get_edgecolor(), \n",
314+
" linewidth=2,\n",
315+
" linestyle=\"dashed\",\n",
316+
" zorder=0, \n",
317+
" )\n",
318+
" transformation = (\n",
319+
" Affine2D()\n",
320+
" .rotate_deg(theta) \n",
321+
" .translate(*m)\n",
322+
" + ax2.transData \n",
323+
" )\n",
324+
" \n",
325+
" current.set_transform(transformation)\n",
326+
" \n",
327+
" ax2.add_patch(current)\n",
328+
" cma.step(sphere)\n",
329+
" \n",
330+
"\n",
331+
"for ax in ax1, ax2:\n",
332+
" ax.set_aspect(\"equal\")\n",
333+
" ax.set_ylim(-5, 0)\n",
334+
" ax.set_xticks([])\n",
335+
" ax.set_yticks([])\n",
336+
"plt.tight_layout()\n",
337+
"plt.savefig(\"figures/adaptation.pdf\")"
338+
]
339+
},
340+
{
341+
"cell_type": "code",
342+
"execution_count": null,
232343
"metadata": {},
233344
"outputs": [],
234345
"source": [
235-
"vector = np.array([0, -1])\n",
346+
"def get_one_plus_one(problem, dim, sampler):\n",
347+
" modules = parameters.Modules()\n",
348+
" modules.sample_transformation = options.SampleTranformerType(sampler)\n",
349+
" \n",
350+
" x0 = np.random.uniform(-5, 5, size=dim)\n",
351+
" return es.OnePlusOneES(\n",
352+
" dim,\n",
353+
" x0=x0,\n",
354+
" f0=problem(x0),\n",
355+
" sigma0=1,\n",
356+
" modules=modules,\n",
357+
" )\n",
358+
" \n",
359+
"names = sorted([options.SampleTranformerType(sampler).name.title().replace(\"Scaled_\", \"\").replace(\"Double_\", \"d\") \n",
360+
" for sampler in range(1, 7)], key=lambda x:x.lower())\n",
361+
"colors = dict(zip(names, sns.color_palette(\"tab10\")))\n",
362+
"\n",
363+
"n_evals = 10000\n",
364+
"n_runs = 1000\n",
365+
"\n",
366+
"linestyle = {\n",
367+
" 2: \"solid\",\n",
368+
" 10: \"dashed\",\n",
369+
" 50: \"dotted\"\n",
370+
"}\n",
371+
"\n",
372+
"np.random.seed(1)\n",
373+
"utils.set_seed(1)\n",
374+
"linewidth = 2\n",
236375
"\n",
237-
"\n"
376+
"f, ax = plt.subplots(figsize=(7, 4))\n",
377+
"for sampler in range(1, 7):\n",
378+
" for d in (2, 10, 50):\n",
379+
" f = np.zeros(n_evals)\n",
380+
" s = np.zeros(n_evals)\n",
381+
" for r in range(n_runs):\n",
382+
" alg = get_one_plus_one(sphere, d, sampler)\n",
383+
" for e in range(n_evals):\n",
384+
" f[e] += alg.f\n",
385+
" s[e] += alg.sigma\n",
386+
" alg.step(sphere)\n",
387+
" \n",
388+
" f /= n_runs\n",
389+
" s /= n_runs\n",
390+
" sampler_name = get_name(alg.sampler.__class__)\n",
391+
" ax.plot(s, color=colors[sampler_name], linestyle=linestyle[d], label=sampler_name, linewidth=linewidth)\n",
392+
"\n",
393+
"ax.legend()\n",
394+
"ax.set_yscale(\"log\")\n",
395+
"ax.set_xscale(\"log\")\n",
396+
"\n",
397+
"from matplotlib.lines import Line2D\n",
398+
"\n",
399+
"handles = [Line2D([0], [0], linestyle='')]\n",
400+
"labels = [\"$\\\\bf{Sampler}$\"]\n",
401+
"\n",
402+
"for alg, color in colors.items():\n",
403+
" handles.append(Line2D([0], [0], label=alg, color=color, linewidth=linewidth))\n",
404+
" labels.append(alg)\n",
405+
"\n",
406+
"\n",
407+
"handles.append(Line2D([0], [0], linestyle=''))\n",
408+
"labels.append(\"$\\\\mathbf{n}$\")\n",
409+
"\n",
410+
"for alg, color in linestyle.items():\n",
411+
" handles.append(Line2D([0], [0], label=alg, color='black', linestyle=color, linewidth=linewidth))\n",
412+
" labels.append(alg)\n",
413+
"\n",
414+
"handles.append(Line2D([0], [0], linestyle=''))\n",
415+
"labels.append(\"\") \n",
416+
"handles.append(Line2D([0], [0], linestyle=''))\n",
417+
"labels.append(\"\") \n",
418+
"\n",
419+
"ax.legend(handles, labels, loc='lower left', fancybox=True, shadow=True, fontsize=13, ncol=2)\n",
420+
"ax.set_ylim(1e-10, 10)\n",
421+
"ax.grid(which=\"both\", axis=\"both\")\n",
422+
"ax.set_ylabel(r\"$\\sigma$\")\n",
423+
"ax.set_xlabel(r\"Evaluations\")\n",
424+
"plt.tight_layout()\n",
425+
"plt.savefig(\"figures/1p1_sigma.pdf\")"
426+
]
427+
},
428+
{
429+
"cell_type": "code",
430+
"execution_count": null,
431+
"metadata": {},
432+
"outputs": [],
433+
"source": [
434+
"samplers = [\n",
435+
" lambda a: stats.cauchy(2.0, scale=1).rvs(size=a),\n",
436+
" lambda a: stats.dweibull(2.0, scale=1).rvs(size=a),\n",
437+
" lambda a: stats.norm().rvs(size=a),\n",
438+
" lambda a: stats.laplace().rvs(size=a),\n",
439+
" lambda a: stats.logistic().rvs(size=a),\n",
440+
" lambda a: stats.uniform().rvs(size=a),\n",
441+
"]\n",
442+
"\n",
443+
"labels=[\"Cauchy\", \"dWeibull\", \"Gaussian\", \"Laplace\", \"Logistic\", \"Uniform\"]\n",
444+
"\n",
445+
"def time_sampler(sampler, n = 1_000_000):\n",
446+
" start = perf_counter()\n",
447+
" sampler(n)\n",
448+
" return perf_counter() - start\n",
449+
"\n",
450+
"t = []\n",
451+
"for label, sampler in zip(labels, samplers):\n",
452+
" times = [time_sampler(sampler) for _ in range(1000)]\n",
453+
" t.append((label, np.mean(times), np.std(times)))\n",
454+
" \n",
455+
"time_data = pl.DataFrame(t, schema=['sampler', 'mean', 'std'], orient='row')"
456+
]
457+
},
458+
{
459+
"cell_type": "code",
460+
"execution_count": null,
461+
"metadata": {},
462+
"outputs": [],
463+
"source": [
464+
"plt.figure(figsize=(6, 3))\n",
465+
"p = plt.errorbar(\n",
466+
" time_data['sampler'], time_data['mean'], time_data['std'], \n",
467+
" marker='_', markersize=20, \n",
468+
" capsize=5,\n",
469+
" markeredgewidth=2, \n",
470+
" elinewidth=2, \n",
471+
" linestyle=''\n",
472+
")\n",
473+
"\n",
474+
"\n",
475+
"plt.grid()\n",
476+
"plt.xticks(rotation=25);\n",
477+
"plt.ylabel(\"Time [s]\", color=p[0].get_color())\n",
478+
"plt.yticks(color=p[0].get_color())\n",
479+
"ax1 = plt.gca()\n",
480+
"ax2 = plt.twinx()\n",
481+
"g_time = time_data.filter(sampler='Gaussian')['mean'] \n",
482+
"\n",
483+
"\n",
484+
"ax2.plot(time_data['sampler'], (time_data['mean'] - g_time) / g_time, color='red', marker='x', linestyle='', markersize=7, markeredgewidth=2)\n",
485+
"\n",
486+
"ax2.set_ylabel(\"vs. Gaussian\", color='red')\n",
487+
"ax2.plot(time_data['sampler'], np.zeros(6), linestyle='dashed', zorder=-100, color='grey')\n",
488+
"plt.yticks(color='red')\n",
489+
"ax2.set_ylim(*((np.array(ax1.get_ylim()) - g_time[0]) / g_time[0]))\n",
490+
"plt.tight_layout()\n",
491+
"plt.savefig(\"figures/time.pdf\")"
238492
]
239493
}
240494
],

‎scripts/distributions/run.sh‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -16,10 +16,8 @@ python run.py --sampler 2 --base_sampler=1 --logged&
1616

1717
for i in {1..6};
1818
do
19-
python run.py --sampler $i --logged --alg 1 &
20-
python run.py --sampler $i --logged --alg 2 &
21-
python run.py --sampler $i --logged --alg 3 &
22-
python run.py --sampler $i --logged --alg 4 &
23-
python run.py --sampler $i --logged --alg 5 &
24-
python run.py --sampler $i --logged --alg 6 &
19+
for j in {1..6};
20+
do
21+
python run.py --sampler $i --logged --alg &j &
22+
done
2523
done

0 commit comments

Comments
 (0)