Word2Vec模型保存与加载的两种方式

model = Word2Vec(common_texts, size=100, window=5, min_count=1, workers=4)

方法一：model.save(path)

• 保存： model.save(path)
• 加载：Word2Vec.load(path)
• 模型可继续训练
  训练时流式的，这意味这句子可以时一个生成器，动态的从磁盘读取输入数据，而无需将整个语料库加载到RAM中。这也意味着我们可以在以后继续训练模型

使用以下命令初始化模型
from gensim.test.utils import common_texts, get_tmpfile
from gensim.models import Word2Vec
 
path = get_tmpfile("word2vec.model") #创建临时文件
 
model = Word2Vec(sentences, size=100, window=5, min_count=1, workers=4)
model.save("word2vec.model")

#加载模型
model = Word2Vec.load("word2vec.model")
#继续训练模型
model.train(["hello","world"], total_examples=1, epochs=1)
#模型调用
vector = model['computer']  # get numpy vector of a word
sims =model.wv.most_similar("computer",topn=10)  #get other similar words
# 如果已经完成了模型的培训(即不再更新，只进行查询)，可以切换到KeyedVectors实例
word_vectors = model.wv
del model  

方法二： model.wv.save(path)

• 保存：model.wv.save(path)
• 加载：wv = KeyedVectors.load("model.wv", mmap='r')
• 模型不可继续训练

from gensim.models import KeyedVectors
word_vectors = model.wv  # Store just the words + their trained embeddings.
word_vectors.save("word2vec.wordvectors")
wv = KeyedVectors.load("word2vec.wordvectors", mmap='r')  # Load back with memory-mapping = read-only, shared across processes.
vector = wv['computer']  # Get numpy vector of a word

link guid:https://radimrehurek.com/gensim/models/word2vec.html