Update notebook

Author: jonasspinner

crates/evaluation/.gitignore

@@ -0,0 +1,3 @@
+copy_from_server.sh
+figures
+data/experiments.csv

crates/evaluation/Evaluation.ipynb

@@ -25,14 +25,34 @@
     "sns.set_theme(context=\"paper\", style=\"white\")"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "73158062-f645-4e9b-9104-8a8850fb5509",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!python3 scripts/combine_data.py"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3daf29a1-0d99-4be0-9e3e-86e8ce05460d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!mkdir -p figures"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
    "id": "2ce88784-56ab-49d1-b79c-875abbde17cc",
    "metadata": {},
    "outputs": [],
    "source": [
-    "df = pd.read_csv(\"../data/experiments.csv\")\n",
+    "df = pd.read_csv(\"./data/experiments.csv\")\n",
     "df[\"n+m\"] = df[\"n\"] + df[\"m\"]\n",
     "df = df[df[\"n+m\"] != 0]\n",
     "df[\"log(n+m)\"] = np.log(df[\"n+m\"])\n",
@@ -70,6 +90,17 @@
     "data.groupby(\"dataset\")[\"time/(n+m) [ns]\"].agg([\"count\", \"mean\"])"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fa7e01da-ffaf-4b1c-a7de-ac182949679c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "data = df.copy()\n",
+    "data.groupby([\"dataset\", \"algo\"], observed=True)[\"time/(n+m) [ns]\"].mean()"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -389,14 +420,6 @@
     "plt.show()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "fd47d899-9156-48d4-9417-640301778a04",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
   {
    "cell_type": "markdown",
    "id": "6969b204-1770-4462-9ad9-91d7a292fbcc",
@@ -503,7 +526,7 @@
    "id": "d71985d0-2a3f-4f46-8c62-11db9899e15d",
    "metadata": {},
    "source": [
-    "# Edge cases"
+    "# Simple graphs"
    ]
   },
   {
@@ -513,22 +536,105 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "fig, axes = plt.subplots(figsize=(6, 2.5), ncols=3, sharey=True)\n",
+    "fig, axes = plt.subplots(figsize=(6, 3), ncols=3, sharey=True)\n",
     "for ax, dataset, title in zip(axes, (\"empty\", \"path\", \"cycle\"), (\"Empty graphs $E_n$\", \"Path graphs $P_n$\", \"Cycle graphs $C_n$\")):\n",
     "    data = df[df[\"dataset\"] == dataset]\n",
     "    ax.set(title=title)\n",
     "    ax.set(ylim=(0, 3.5), axisbelow=True)\n",
     "    ax.tick_params(axis='x', which='major', bottom=True, labelsize=8)\n",
     "    ax.grid(True)\n",
-    "    sns.scatterplot(x=\"n\", y=\"time/n [μs]\", hue=\"algo\", edgecolor=None, s=2, data=data, ax=ax, rasterized=True)\n",
-    "for ax in axes[:-1]:\n",
+    "    sns.scatterplot(x=\"n+m\", y=\"time/(n+m) [μs]\", hue=\"algo\", edgecolor=None, s=2, data=data, ax=ax, rasterized=True)\n",
     "    ax.legend([],[], frameon=False)\n",
-    "axes[-1].legend(bbox_to_anchor=(1.0, 1.0), frameon=False, markerscale=4, title=\"Algorithm\")\n",
+    "axes[1].legend(bbox_to_anchor=(1.0, 1.0), markerscale=4, title=\"Algorithm\")\n",
     "sns.despine(left=True)\n",
     "fig.tight_layout()\n",
     "plt.savefig(\"figures/empty-path-cycle.pdf\", dpi=300)\n",
     "plt.show()"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "605bdb3a-1b34-45a7-931d-66df1276c486",
+   "metadata": {},
+   "source": [
+    "# Anaylsis of $G(n, m)$ model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "af30f324-dbe9-4914-9764-d45e93403a55",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import networkx as nx\n",
+    "from tqdm.contrib.concurrent import process_map\n",
+    "from itertools import product\n",
+    "from subprocess import run\n",
+    "from tempfile import NamedTemporaryFile, TemporaryDirectory\n",
+    "\n",
+    "import sys  \n",
+    "sys.path.insert(1, '../scripts')\n",
+    "import analyze\n",
+    "\n",
+    "def write_metis_to_file(f, graph: nx.Graph):\n",
+    "    graph = nx.convert_node_labels_to_integers(graph)\n",
+    "    f.write(f\"{graph.number_of_nodes()} {graph.number_of_edges()}\\n\".encode())\n",
+    "    for u in graph.nodes:\n",
+    "        f.write((\" \".join(str(v+1) for v in graph[u]) + \"\\n\").encode())\n",
+    "    f.flush()\n",
+    "    f.seek(0)\n",
+    "\n",
+    "def modular_decomposition(graph: nx.Graph):\n",
+    "    with NamedTemporaryFile() as input_file, TemporaryDirectory() as dir:\n",
+    "        write_metis_to_file(input_file, graph)\n",
+    "        output = Path(dir) / \"out\"\n",
+    "        cmd = [\"../target/release/md\", \"--input-type\", \"metis\", \"--input\", input_file.name, \"--algo\", \"fracture\", \"--output\", output]\n",
+    "        out = run(cmd, capture_output=True)\n",
+    "        out.check_returncode()\n",
+    "        return analyze.analyze_tree(output, only_header=False, timeout=10)\n",
+    "\n",
+    "def generate_data(params):\n",
+    "    n, m, seed = params\n",
+    "    m = int(m)\n",
+    "    graph = nx.gnm_random_graph(n, m, seed=seed)\n",
+    "    res = modular_decomposition(graph)\n",
+    "    return [dict(n=n, m=m, seed=seed, kind=kind, num=num) for kind, num in zip((\"prime\", \"series\", \"parallel\"), map(int, res.split(\",\")[5:8]))]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "11b6cbbe-9824-47c6-a39d-9e3601782aa3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "if False:\n",
+    "    params = [(n, m, seed) for n, repeats in [(2**10, 40), (2**16, 10)] for m, seed in product(np.linspace(0, 3 * 2**10, 81), range(repeats))]\n",
+    "    rows = process_map(generate_data, params, chunksize=16)\n",
+    "else:\n",
+    "    rows = []"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "c302fa6e-3eb7-4cdf-9e93-699a7e7c4922",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "data = pd.DataFrame([row for group in rows for row in group])\n",
+    "data = data[data[\"n\"].isin([2**10, 2**16])]\n",
+    "data[\"m/n\"] = data[\"m\"] / data[\"n\"]\n",
+    "data[\"num/n\"] = data[\"num\"] / data[\"n\"]\n",
+    "\n",
+    "fig, ax = plt.subplots(figsize=(6, 2.5))\n",
+    "sns.lineplot(x=\"m/n\", y=\"num/n\", hue=\"kind\", style=\"n\", rasterized=True, data=data, ax=ax)\n",
+    "ax.legend(bbox_to_anchor=(1, 1), loc=\"upper left\", frameon=False)\n",
+    "sns.despine()\n",
+    "plt.savefig(\"figures/gnm-module-distribution.pdf\", dpi=300)\n",
+    "plt.show()"
+   ]
   }
  ],
  "metadata": {