{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": { "collapsed": false }, "outputs": [], "source": [ "%matplotlib inline" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n# Module to run a single or multiple examples\n\nModule to run the benchmark across all the baseline embedding algorithms.\n\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "collapsed": false }, "outputs": [], "source": [ "from subprocess import call\nimport itertools\ntry: import cPickle as pickle\nexcept: import pickle\nimport json\nimport networkx as nx\nimport pandas as pd\nimport pdb\nimport os\nimport sys\nfrom time import time\nfrom gemben.utils import graph_gens\n\nmethClassMap = {\"gf\": \"GraphFactorization\",\n \"hope\": \"HOPE\",\n \"lap\": \"LaplacianEigenmaps\",\n \"node2vec\": \"node2vec\",\n \"sdne\": \"SDNE\",\n \"pa\": \"PreferentialAttachment\",\n \"rand\": \"RandomEmb\",\n \"cn\": \"CommonNeighbors\",\n \"aa\": \"AdamicAdar\",\n \"jc\": \"JaccardCoefficient\"}\n\nclass exp:\n def __init__(self, domain=\"social\", method=\"sdne\", rounds=1,lexp=False, samp_scheme='rw', plot_hyp_data=False):\n \n t1 = time()\n self.params = json.load(\n open('gemben/experiments/config/params_benchmark.conf', 'r')\n )\n self.domain_graph_map = json.load(\n open('gemben/experiments/config/domain_graph_map.conf', 'r')\n )\n # graph_hyp_range: {N: [128, 256, 512, 1024], deg: [4, 6,8, 10, 12]}\n self.graph_hyp_range = json.load(\n open('gemben/experiments/config/graph_hyp_range.conf', 'r')\n )\n # def_graph_hyps: {N: 1024, deg: 8, dia: None, dim: 128}\n self.def_graph_hyps = json.load(\n open('gemben/experiments/config/def_graph_hyps.conf', 'r')\n )\n\n self.params[\"rounds\"] = rounds\n self.params[\"graphs\"] = self.domain_graph_map[domain]\n self.params[\"lexp\"] = lexp\n self.params[\"plot_hyp_data\"] = plot_hyp_data\n if method == \"all\":\n self.params[\"methods\"] = methClassMap.keys()\n elif len(method)>1:\n self.params[\"methods\"] = method.split(',')\n else:\n self.params[\"methods\"] = self.method\n self.samp_scheme = samp_scheme\n\n\n def run(self): \n\n try:\n os.makedirs(\"gemben/intermediate\")\n except:\n pass\n try:\n os.makedirs(\"gemben/results\")\n except:\n pass\n try:\n os.makedirs(\"gemben/temp\")\n except:\n pass\n\n graph_hyp_keys = list(self.graph_hyp_range.keys())\n ev_cols = [\"LP MAP\", \"LP P@100\"]\n for meth , graph in itertools.product(*[self.params[\"methods\"],self.params[\"graphs\"]]):\n hyp_df = pd.DataFrame(\n columns=graph_hyp_keys + ev_cols + [\"Round Id\"]\n )\n hyp_r_idx = 0\n for hyp_key in graph_hyp_keys:\n \n for curr_hyp_key_range, r_id in itertools.product(\n *[graph_hyp_range[hyp_key], range(self.params[\"rounds\"])]\n ):\n \n if r_id == 0: \n f_hyp = 1\n else:\n f_hyp = 0\n \n curr_hyps = self.def_graph_hyps.copy()\n \n curr_hyps[hyp_key] = curr_hyp_key_range\n curr_hyps[\"domain\"] = self.params[\"domain_name\"]\n hyp_str = '_'.join(\n \"%s=%s\" % (key, str(val).strip(\"'\")) for (key, val) in curr_hyps.items()\n )\n \n hyp_str_graph_name = '_'.join(\n \"%s=%s\" % (key, str(val).strip(\"'\")) for (key, val) in curr_hyps.items() if key != 'dim' \n )\n \n syn_data_folder = 'benchmark_%s_%s_%s' % (graph, hyp_str_graph_name, r_id)\n \n graphClass = getattr(graph_gens, graph)\n \n try:\n nx.read_gpickle(\n 'gemben/data/%s/graph.gpickle' % syn_data_folder\n )\n except:\n flag = 1\n ##### flag = 0 means the labels are continous on lcc\n while flag:\n print(\"Graph is generating...\")\n G = graphClass(**curr_hyps)[0]\n if len(set(G.nodes())) == G.number_of_nodes() and list(G.nodes())[-1] == G.number_of_nodes() -1:\n flag = 0\n if G:\n if not os.path.exists(\"gemben/data/%s\" % syn_data_folder):\n os.makedirs(\"gemben/data/%s\" % syn_data_folder)\n nx.write_gpickle(\n G, 'gemben/data/%s/graph.gpickle' % syn_data_folder\n )\n perf_exp = not self.params[\"lexp\"] \n if self.params[\"lexp\"]:\n try:\n MAP, prec, n_samps = pickle.load(\n open('gemben/results/%s_%s_%d_%s.lp' % (\n syn_data_folder, meth, \n curr_hyps[\"dim\"], self.samp_scheme), 'rb'))\n except: \n perf_exp = 1\n ##### only find the best hyp for first round\n if perf_exp:\n os.system(\n \"python3 gemben/experiments/exp.py -data %s -meth %s -dim %d -rounds 1 -find_hyp %d -s_sch %s -exp lp\" % (\n syn_data_folder,\n meth,\n curr_hyps[\"dim\"],\n f_hyp,\n self.samp_scheme\n )\n )\n MAP, prec, n_samps = pickle.load(\n open('gemben/results/%s_%s_%d_%s.lp' % (\n syn_data_folder, meth, \n curr_hyps[\"dim\"], self.samp_scheme), 'rb')) \n hyp_df.loc[hyp_r_idx, graph_hyp_keys] = \\\n pd.Series(curr_hyps)\n #prec_100 = prec[int(n_samps[0])][0][100]\n try:\n prec_100 = list(prec.values())[0][0][100]\n except:\n pdb.set_trace()\n f_temp = open(\"gemben/temp/%s_%s_%s_lp_%s_data_hyp.txt\" % (\n self.params[\"domain_name\"], graph, meth, self.samp_scheme), 'a')\n f_temp.write('%s: round: %d, MAP: %f, prec_100: %f' % (hyp_str, r_id, list(MAP.values())[0][0], prec_100))\n f_temp.close()\n hyp_df.loc[hyp_r_idx, ev_cols + [\"Round Id\"]] = \\\n [list(MAP.values())[0][0], prec_100, r_id]\n #[MAP[int(n_samps[0])][0], prec_100, r_id]\n hyp_r_idx += 1\n\n hyp_df.to_hdf(\n \"gemben/intermediate/%s_%s_%s_lp_%s_data_hyp.h5\" % (\n self.params[\"domain_name\"], graph, meth, self.samp_scheme),\n \"df\"\n )\n print('Experiments done for %s, %s' % (graph, meth))" ] } ], "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": "3.6.8" } }, "nbformat": 4, "nbformat_minor": 0 }