| 24 | |
| 25 | |
| 26 | class Node2Vec: |
| 27 | |
| 28 | def __init__(self, graph, walk_length, num_walks, p=1.0, q=1.0, workers=1, use_rejection_sampling=False): |
| 29 | |
| 30 | self.graph = graph |
| 31 | self._embeddings = {} |
| 32 | self.walker = RandomWalker( |
| 33 | graph, p=p, q=q, use_rejection_sampling=use_rejection_sampling) |
| 34 | |
| 35 | print("Preprocess transition probs...") |
| 36 | self.walker.preprocess_transition_probs() |
| 37 | |
| 38 | self.sentences = self.walker.simulate_walks( |
| 39 | num_walks=num_walks, walk_length=walk_length, workers=workers, verbose=1) |
| 40 | |
| 41 | def train(self, embed_size=128, window_size=5, workers=3, iter=5, **kwargs): |
| 42 | kwargs["sentences"] = self.sentences |
| 43 | kwargs["min_count"] = kwargs.get("min_count", 0) |
| 44 | kwargs["vector_size"] = embed_size |
| 45 | kwargs["sg"] = 1 |
| 46 | kwargs["hs"] = 0 # node2vec not use Hierarchical Softmax |
| 47 | kwargs["workers"] = workers |
| 48 | kwargs["window"] = window_size |
| 49 | kwargs["epochs"] = iter |
| 50 | |
| 51 | print("Learning embedding vectors...") |
| 52 | model = Word2Vec(**kwargs) |
| 53 | print("Learning embedding vectors done!") |
| 54 | |
| 55 | self.w2v_model = model |
| 56 | |
| 57 | return model |
| 58 | |
| 59 | def get_embeddings(self, ): |
| 60 | if self.w2v_model is None: |
| 61 | print("model not train") |
| 62 | return {} |
| 63 | |
| 64 | self._embeddings = {} |
| 65 | for word in self.graph.nodes(): |
| 66 | self._embeddings[word] = self.w2v_model.wv[word] |
| 67 | |
| 68 | return self._embeddings |
no outgoing calls