diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..4794c40 --- /dev/null +++ b/.gitignore @@ -0,0 +1,112 @@ +# Byte-compiled / optimized / DLL files +__pycache__/ +*.py[cod] +*$py.class + +# C extensions +*.so + +# Distribution / packaging +.Python +build/ +develop-eggs/ +dist/ +downloads/ +eggs/ +.eggs/ +lib/ +lib64/ +parts/ +sdist/ +var/ +wheels/ +*.egg-info/ +.installed.cfg +*.egg +MANIFEST + +# PyInstaller +# Usually these files are written by a python script from a template +# before PyInstaller builds the exe, so as to inject date/other infos into it. +*.manifest +*.spec + +# Installer logs +pip-log.txt +pip-delete-this-directory.txt + +# Unit test / coverage reports +htmlcov/ +.tox/ +.coverage +.coverage.* +.cache +nosetests.xml +coverage.xml +*.cover +.hypothesis/ +.pytest_cache/ + +# Translations +*.mo +*.pot + +# Django stuff: +*.log +local_settings.py +db.sqlite3 + +# Flask stuff: +instance/ +.webassets-cache + +# Scrapy stuff: +.scrapy + +# Sphinx documentation +docs/_build/ + +# PyBuilder +target/ + +# Jupyter Notebook +.ipynb_checkpoints + +# pyenv +.python-version + +# celery beat schedule file +celerybeat-schedule + +# SageMath parsed files +*.sage.py + +# Environments +.env +.venv +env/ +venv/ +ENV/ +env.bak/ +venv.bak/ + +# Spyder project settings +.spyderproject +.spyproject + +# Rope project settings +.ropeproject + +# mkdocs documentation +/site + +# mypy +.mypy_cache/ + +# pycharm +.idea +.idea/ + +# data folder +data/ +data diff --git a/LICENSE b/LICENSE new file mode 100644 index 0000000..bdf1a0b --- /dev/null +++ b/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2019 Anirban Biswas + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/README.md b/README.md index 7a3235f..d1f4f58 100644 --- a/README.md +++ b/README.md @@ -1,2 +1,19 @@ -# mirand -Information network representation using structure and content +## How to run - Step by step + +1. Use generate_content_edgelist.ipynb in preprocessing folder to generate the cosine_content.edgelist file. This is basically the graph generated from the content. Change the dataset name accordingly. Vary the theta parameter if required. By default, theta = 1 +2. Use generate_two_level_embed_from_structure_and_content.ipynb in embedding_generation folder to generate the two level embedding. Change the dataset name accordingly. The embedding file is generated in the corresponding dataset folder in data directory. The default dimension is 128. Change the dimension parameter if you wish to generate embeddings of some other dimension. +3. The previously mentioned ipython file also converts the embedding file to a suitable csv format required for downstream data mining tasks. +4. Use the sac2vec_tasks.ipynb in tasks folder for classification, clustering tasks and visualization tasks. Change the dataset name accordingly. + +## Data folder + +- Look st the sample dataset cora. If you want to experiment on different datasets, create a folder with name of your dataset. +- Three files are required to run all the steps - content.csv, label.csv and reference.edgelist +- reference.edgelist file consists of the graph structure in edgelist format. +- content.csv file is content associated with the nodes. +- label.csv contains the labels of the nodes. This file is required for classification and clustering. + + +## Sample command to run + +python main.py --input-struc ../data/cora/cora_struc.edgelist --input-attr ../data/cora/cora_attr.edgelist --output ../data/cora/cora.embed --dataset=cora --dimensions=8 \ No newline at end of file diff --git a/requirements.txt b/requirements.txt new file mode 100644 index 0000000..f294334 --- /dev/null +++ b/requirements.txt @@ -0,0 +1,28 @@ +backports.functools-lru-cache==1.5 +boto==2.49.0 +boto3==1.9.71 +botocore==1.12.71 +bz2file==0.98 +certifi==2018.11.29 +chardet==3.0.4 +cycler==0.10.0 +decorator==4.3.0 +docutils==0.14 +futures==3.2.0 +gensim==0.13.3 +idna==2.8 +jmespath==0.9.3 +kiwisolver==1.0.1 +matplotlib==2.2.3 +networkx==1.11 +numpy==1.11.2 +pyparsing==2.3.1 +python-dateutil==2.7.5 +pytz==2018.9 +requests==2.21.0 +s3transfer==0.1.13 +scipy==1.2.0 +six==1.12.0 +smart-open==1.7.1 +subprocess32==3.5.3 +urllib3==1.24.1 diff --git a/src/main.py b/src/main.py new file mode 100644 index 0000000..348e431 --- /dev/null +++ b/src/main.py @@ -0,0 +1,145 @@ +import argparse +import networkx as nx +import mirand +from gensim.models import Word2Vec + + +def parse_args(): + ''' + Parses the MIRand arguments. + ''' + parser = argparse.ArgumentParser(description="Run MIRand.") + + parser.add_argument('--input-struc', nargs='?', default='graph/karate.edgelist', + help='Input graph path for structure-layer') + + parser.add_argument('--input-attr', nargs='?', default='graph/karate.edgelist', + help='Input graph path for attributes-layer') + + parser.add_argument('--dataset', nargs='?', default='karate', + help='Input graph name for saving files') + + parser.add_argument('--output', nargs='?', default='emb/karate.emb', + help='Embeddings path') + + parser.add_argument('--dimensions', type=int, default=128, + help='Number of dimensions. Default is 128.') + + parser.add_argument('--walk-length', type=int, default=80, + help='Length of walk per source. Default is 80.') + + parser.add_argument('--num-walks', type=int, default=10, + help='Number of walks per source. Default is 10.') + + parser.add_argument('--window-size', type=int, default=10, + help='Context size for optimization. Default is 10.') + + parser.add_argument('--iter', default=1, type=int, + help='Number of epochs in SGD') + + parser.add_argument('--workers', type=int, default=8, + help='Number of parallel workers. Default is 8.') + + parser.add_argument('--p-struc', type=float, default=1, + help='Return hyperparameter for Structure-Layer. Default is 1.') + + parser.add_argument('--q-struc', type=float, default=1, + help='Inout hyperparameter for Structure-Layer. Default is 1.') + + parser.add_argument('--p-attr', type=float, default=1, + help='Return hyperparameter for Attribute-Layer. Default is 1.') + + parser.add_argument('--q-attr', type=float, default=1, + help='Inout hyperparameter for Attribute-Layer. Default is 1.') + + parser.add_argument('--weighted-struc', dest='weighted', action='store_true', + help='Boolean specifying (un)weighted Structure-Layer. Default is unweighted.') + parser.add_argument('--unweighted-struc', dest='unweighted', action='store_false') + parser.set_defaults(weighted_struc=False) + + parser.add_argument('--weighted-attr', dest='weighted', action='store_true', + help='Boolean specifying (un)weighted Attribute-Layer. Default is weighted.') + parser.add_argument('--unweighted-attr', dest='unweighted', action='store_false') + parser.set_defaults(weighted_attr=True) + + parser.add_argument('--directed-struc', dest='directed', action='store_true', + help='Structure-Layer Graph is (un)directed. Default is undirected.') + parser.add_argument('--undirected-struc', dest='undirected', action='store_false') + parser.set_defaults(directed_struc=False) + + parser.add_argument('--directed-attr', dest='directed', action='store_true', + help='Attribute-Layer Graph is (un)directed. Default is directed.') + parser.add_argument('--undirected-attr', dest='undirected', action='store_false') + parser.set_defaults(directed_attr=True) + + return parser.parse_args() + + +def read_graph(): + ''' + Reads the structure and attribute network in networkx. + ''' + if args.weighted_attr: + A = nx.read_edgelist(args.input_attr, nodetype=int, data=(('weight', float),), create_using=nx.DiGraph()) + else: + A = nx.read_edgelist(args.input_attr, nodetype=int, create_using=nx.DiGraph()) + for edge in A.edges(): + A[edge[0]][edge[1]]['weight'] = 1 + + if args.weighted_struc: + S = nx.read_edgelist(args.input_struc, nodetype=int, data=(('weight', float),), create_using=nx.DiGraph()) + else: + S = nx.read_edgelist(args.input_struc, nodetype=int, create_using=nx.DiGraph()) + for edge in S.edges(): + S[edge[0]][edge[1]]['weight'] = 1 + + if not args.directed_struc: + S = S.to_undirected() + + if not args.directed_attr: + A = A.to_undirected() + + return S, A + + +def generate_walks(G, start=False): + ################################# + # Calculate transition probabilities for switching the levels + print('\nStructure layer trans-prob evaluation started') + G.get_level_transition_weight(0) + print('\nAttribute layer trans-prob evaluation started') + G.get_level_transition_weight(1) + + # print('\nalias 1 started') + G.preprocess_transition_probs(0) + # print('\nalias 2 started') + G.preprocess_transition_probs(1) + + print('\nWalk simulation started') + walks = G.simulate_walks(args.num_walks, args.walk_length) + return walks + + +def learn_embeddings(G): + walks = [map(str, walk) for walk in generate_walks(G, start=True)] + model = Word2Vec(walks, size=args.dimensions, window=args.window_size, min_count=0, sg=1, workers=args.workers, iter=args.iter) + model.save_word2vec_format(args.output) + return + + +def main(args): + print('Running MIRand for %s dataset\n' % args.dataset) + nx_S, nx_A = read_graph() + G = mirand.Graph(nx_S, nx_A, args.directed_struc, args.directed_attr, args.p_struc, args.q_struc, args.p_attr, args.q_attr) + + assert len(nx_S.nodes()) == len(nx_A.nodes()) + print("Number of nodes in the graph : %s" % len(nx_S.nodes())) + print("Number of edges in the structure graph : %s" % len(nx_S.edges())) + print("Number of edges in the attribute graph : %s" % len(nx_A.edges())) + + learn_embeddings(G) + + +if __name__ == "__main__": + args = parse_args() + main(args) diff --git a/src/mirand.py b/src/mirand.py new file mode 100644 index 0000000..84fd6cd --- /dev/null +++ b/src/mirand.py @@ -0,0 +1,184 @@ +import numpy as np +import networkx as nx +import random + + +class Graph(): + def __init__(self, nx_S, nx_A, is_directed_struc, is_directed_attr, p_struc, q_struc, p_attr, q_attr): + self.G = [nx_S, nx_A] + self.is_directed = [is_directed_struc, is_directed_attr] + self.p = [p_struc, p_attr] + self.q = [q_struc, q_attr] + self.alias_nodes = [None, None] + self.alias_edges = [None, None] + self.trans_weight = [None, None] + self.ct = [0, 0] + + # Modified node2vec walk for multi-layered graph with structure and content + def node2vec_walk(self, walk_length, start_node): + ''' + Simulate a random walk starting from start node. + ''' + alias_nodes = self.alias_nodes + alias_edges = self.alias_edges + + walk = [start_node] + while len(walk) < walk_length: + cur = walk[-1] + # Decide the layer structure/attribute for the current node + # taking upper as structure and lower level as attribute + up = self.trans_weight[1][cur] + down = self.trans_weight[0][cur] + pu = up / (up + down) # probability to select in structure layer + pd = 1 - pu # probability to select in attribute layer + + x = random.random() # random num between 0---1 + if x < pu: # if pu is large then more chances of Reference being selected + ind = 0 + else: + ind = 1 + self.ct[ind] += 1 # to get count which layer the Random walk is in. + + cur_nbrs = sorted(self.G[ind].neighbors(cur)) + if len(cur_nbrs) > 0: + if len(walk) == 1: + walk.append(cur_nbrs[alias_draw(alias_nodes[ind][cur][0], alias_nodes[ind][cur][1])]) + else: + prev = walk[-2] + if (prev, cur) not in alias_edges[ind]: # when the edge is not in other graph + walk.append(cur_nbrs[alias_draw(alias_nodes[ind][cur][0], alias_nodes[ind][cur][1])]) + else: + e1 = alias_edges[ind][(prev, cur)][0] + e2 = alias_edges[ind][(prev, cur)][1] + tmp = alias_draw(e1, e2) + next = cur_nbrs[tmp] + walk.append(next) + else: + break + + return walk + + def simulate_walks(self, num_walks, walk_length): + G = self.G[0] # we can take any graph as we just need to find the nodes + walks = [] + nodes = list(G.nodes()) + for walk_iter in range(num_walks): + print('Walk iteration :: %s / %s' % (str(walk_iter + 1), str(num_walks))) + random.shuffle(nodes) + for node in nodes: + walks.append(self.node2vec_walk(walk_length=walk_length, start_node=node)) + print('Walk count in Structure layer :: %s , Attribute layer :: %s' % (str(self.ct[0]), str(self.ct[1]))) + self.ct = [0, 0] + return walks + + def get_level_transition_weight(self, ind): + G = self.G[ind] + mat = nx.to_scipy_sparse_matrix(G) + if ind == 0: + avg = 1.0 + else: + avg = 1.0 * np.sum(mat) / G.number_of_edges() + if ind == 0: + print('Threshold for structure layer :: %s' % str(avg)) + else: + print('Threshold for attribute layer :: %s' % str(avg)) + mat = mat >= avg + tau = np.sum(mat, axis=1) + self.trans_weight[ind] = np.log(np.e + tau) + + def get_alias_edge(self, src, dst, ind): + ''' + Get the alias edge setup lists for a given edge. + ''' + G = self.G[ind] + p = self.p[ind] + q = self.q[ind] + + unnormalized_probs = [] + for dst_nbr in sorted(G.neighbors(dst)): + if dst_nbr == src: + unnormalized_probs.append(G[dst][dst_nbr]['weight'] / p) + elif G.has_edge(dst_nbr, src): + unnormalized_probs.append(G[dst][dst_nbr]['weight']) + else: + unnormalized_probs.append(G[dst][dst_nbr]['weight'] / q) + norm_const = sum(unnormalized_probs) + normalized_probs = [float(u_prob) / norm_const for u_prob in unnormalized_probs] + + return alias_setup(normalized_probs) + + def preprocess_transition_probs(self, ind): + ''' + Preprocessing of transition probabilities for guiding the random walks. + ''' + G = self.G[ind] + is_directed = self.is_directed[ind] + + alias_nodes = {} + for node in G.nodes(): + unnormalized_probs = [G[node][nbr]['weight'] for nbr in sorted(G.neighbors(node))] + norm_const = sum(unnormalized_probs) + normalized_probs = [float(u_prob) / norm_const for u_prob in unnormalized_probs] + alias_nodes[node] = alias_setup(normalized_probs) + + alias_edges = {} + triads = {} + + if is_directed: + for edge in G.edges(): + alias_edges[edge] = self.get_alias_edge(edge[0], edge[1], ind) + else: + for edge in G.edges(): + alias_edges[edge] = self.get_alias_edge(edge[0], edge[1], ind) + alias_edges[(edge[1], edge[0])] = self.get_alias_edge(edge[1], edge[0], ind) + + self.alias_nodes[ind] = alias_nodes + self.alias_edges[ind] = alias_edges + + return + + +def alias_setup(probs): + ''' + Compute utility lists for non-uniform sampling from discrete distributions. + Refer to https://hips.seas.harvard.edu/blog/2013/03/03/the-alias-method-efficient-sampling-with-many-discrete-outcomes/ + for details + ''' + K = len(probs) + q = np.zeros(K) + J = np.zeros(K, dtype=np.int) + + smaller = [] + larger = [] + for kk, prob in enumerate(probs): + q[kk] = K * prob + if q[kk] < 1.0: + smaller.append(kk) + else: + larger.append(kk) + + while len(smaller) > 0 and len(larger) > 0: + small = smaller.pop() + large = larger.pop() + + J[small] = large + q[large] = q[large] + q[small] - 1.0 + if q[large] < 1.0: + smaller.append(large) + else: + larger.append(large) + + return J, q + + +def alias_draw(J, q): + ''' + Draw sample from a non-uniform discrete distribution using alias sampling. + ''' + K = len(J) + kk = int(np.floor(np.random.rand() * K)) + tmp = q[kk] + if np.random.rand() < q[kk]: + return kk + else: + return J[kk]