From c564d38b23809be49670dd59a5e45a44c5ced2ba Mon Sep 17 00:00:00 2001
From: Bas Nijholt <basnijholt@gmail.com>
Date: Mon, 23 Sep 2019 20:39:46 +0200
Subject: [PATCH] improve fig loss error
---
figures.ipynb | 78 +++++++++++++++++++++++++++------------------------
1 file changed, 42 insertions(+), 36 deletions(-)
diff --git a/figures.ipynb b/figures.ipynb
index 063bea0..8187dba 100644
--- a/figures.ipynb
+++ b/figures.ipynb
@@ -22,6 +22,7 @@
"source": [
"import functools\n",
"import itertools\n",
+ "import os\n",
"import pickle\n",
"\n",
"import holoviews.plotting.mpl\n",
@@ -146,9 +147,9 @@
" bounds=phase_diagram_setup(\"phase_diagram.pickle\")[1],\n",
" npoints=100,\n",
" fname=\"phase_diagram.pickle\",\n",
- " title=\"PD\",\n",
+ " title=\"phase diagram\",\n",
" ),\n",
- " dict(function=level_crossing, bounds=[(-1, 1), (-3, 3)], npoints=50, title=\"LC\"),\n",
+ " dict(function=level_crossing, bounds=[(-1, 1), (-3, 3)], npoints=50, title=\"level crossing\"),\n",
"]"
]
},
@@ -631,8 +632,8 @@
"def get_err_2D(learner, N):\n",
" xys_rand = np.vstack(\n",
" [\n",
- " np.random.uniform(*learner.bounds[0], size=int(100_000 ** 0.5)),\n",
- " np.random.uniform(*learner.bounds[1], size=int(100_000 ** 0.5)),\n",
+ " np.random.uniform(*learner.bounds[0], size=int(100_000)),\n",
+ " np.random.uniform(*learner.bounds[1], size=int(100_000)),\n",
" ]\n",
" )\n",
"\n",
@@ -665,33 +666,36 @@
" return err_lin, err_learner\n",
"\n",
"\n",
+ "recalculate = False\n",
"N_max = 10_000\n",
"Ns = np.geomspace(4, N_max, 50).astype(int)\n",
- "\n",
- "loss_1D = adaptive.learner.learner1D.curvature_loss_function()\n",
- "loss_2D = adaptive.learner.learnerND.curvature_loss_function()\n",
- "\n",
- "err = collections.defaultdict(dict)\n",
- "for i, (funcs, loss, Learner, get_err) in enumerate(\n",
- " zip(\n",
- " [funcs_1D, funcs_2D],\n",
- " [loss_1D, loss_2D],\n",
- " [Learner1D, LearnerND],\n",
- " [get_err_1D, get_err_2D],\n",
- " )\n",
- "):\n",
- " for d in funcs:\n",
- " learner = Learner(d[\"function\"], d[\"bounds\"], loss)\n",
- " adaptive.runner.simple(learner, goal=lambda l: l.npoints >= N_max)\n",
- " errs = [get_err(learner, N) for N in Ns]\n",
- " err_hom, err_adaptive = zip(*errs)\n",
- " err[i][d[\"title\"]] = (err_hom, err_adaptive)\n",
- "\n",
- " d[\"err_hom\"] = err_hom\n",
- " d[\"err_adaptive\"] = err_adaptive\n",
- "\n",
- "with open(\"error_line_loss.pickle\", \"wb\") as f:\n",
- " pickle.dump(err, f)"
+ "fname = \"error_line_loss.pickle\"\n",
+ "\n",
+ "if not os.path.exists(fname) and not recalculate:\n",
+ " loss_1D = adaptive.learner.learner1D.curvature_loss_function()\n",
+ " loss_2D = adaptive.learner.learnerND.curvature_loss_function()\n",
+ "\n",
+ " err = collections.defaultdict(dict)\n",
+ " for i, (funcs, loss, Learner, get_err) in enumerate(\n",
+ " zip(\n",
+ " [funcs_1D, funcs_2D],\n",
+ " [loss_1D, loss_2D],\n",
+ " [Learner1D, LearnerND],\n",
+ " [get_err_1D, get_err_2D],\n",
+ " )\n",
+ " ):\n",
+ " for d in funcs:\n",
+ " learner = Learner(d[\"function\"], d[\"bounds\"], loss)\n",
+ " adaptive.runner.simple(learner, goal=lambda l: l.npoints >= N_max)\n",
+ " errs = [get_err(learner, N) for N in Ns]\n",
+ " err_hom, err_adaptive = zip(*errs)\n",
+ " err[i][d[\"title\"]] = (err_hom, err_adaptive)\n",
+ "\n",
+ " with open(fname, \"wb\") as f:\n",
+ " pickle.dump(err, f)\n",
+ "else:\n",
+ " with open(fname, \"rb\") as f:\n",
+ " err = pickle.load(f)"
]
},
{
@@ -700,10 +704,6 @@
"metadata": {},
"outputs": [],
"source": [
- "import pickle, os\n",
- "with open('error_line_loss.pickle', 'rb') as f:\n",
- " err = pickle.load(f)\n",
- " \n",
"fig, axs = plt.subplots(2, 1, figsize=(fig_width, 1.6 * fig_height))\n",
"plt.subplots_adjust(hspace=0.3)\n",
"\n",
@@ -715,8 +715,14 @@
" for j, (title, (err_hom, err_adaptive)) in enumerate(err[i].items()):\n",
" color = f\"C{j}\"\n",
" label = \"abc\"[j]\n",
- " ax.loglog(Ns, err_hom, ls=\"--\", c=color)\n",
- " ax.loglog(Ns, err_adaptive, label=f\"$\\mathrm{{({label})}}$ {title}\", c=color)\n",
+ " label = f\"$\\mathrm{{({label})}}$ {title}\"\n",
+ "# ax.loglog(Ns, err_hom, ls=\"--\", c=color)\n",
+ "# ax.loglog(Ns, err_adaptive, label=label, c=color)\n",
+ " error = np.array(err_hom) / np.array(err_adaptive)\n",
+ " if i == 0:\n",
+ " ax.loglog(Ns[:36], error[:36], c=color, label=label)\n",
+ " else:\n",
+ " ax.loglog(Ns, error, c=color, label=label)\n",
" ax.legend()\n",
"\n",
"plt.savefig(\"figures/line_loss_error.pdf\", bbox_inches=\"tight\", transparent=True)\n",
@@ -734,7 +740,7 @@
"for title, (err_hom, err_adaptive) in err[0].items():\n",
" print(title, err_hom[-1] / err_adaptive[-1])\n",
"\n",
- "print(\"2D\")\n",
+ "print(\"\\n2D\")\n",
"for title, (err_hom, err_adaptive) in err[1].items():\n",
" print(title, err_hom[-1] / err_adaptive[-1])"
]
--
GitLab