Python Gensim从向量创建Word2Vec模型（在ndarray中）

Question

I have a ndarray with words and their corresponding vector (with the size of 100 per word). For example:

Computer 0.11 0.41 ... 0.56
Ball     0.31 0.87 ... 0.32

And so on.

I want to create a word2vec model from it:

model = load_from_ndarray(arr)

How can it be done? I saw

KeyedVectors

but it only takes file and not array

Answer 1

There's no existing convenience methods to turn your own array/word-list into a KeyedVectors . So you'd have to hand-construct that, in your own code.

But it's a pretty simple object, mainly one raw array and a dict for mapping words to index-locations, and all the source is available:

https://github.com/RaRe-Technologies/gensim/blob/develop/gensim/models/keyedvectors.py

I would especially suggest strategies of doing one or both of:

• taking a close look at the load_word2vec_format() method including the similarly-named supporting function in the sibling base_any2vec.py file, and seeing each of the steps they use in reading a file and constructing a full instance

• training up a dummy KeyedVectors in one of the supported ways – such as by training Word2Vec on some synthetic corpus that includes exactly the words you need – and then either inspecting that object to understand the necessary parts of a working instance, or mutating that instance in-place to then have the vector-mappings you prefer.

Answer 2

from gensim.models import KeyedVectors
words = myarray[:,0]
vectors = myarray[:,1:]
model = KeyedVectors(vectors.shape[1])
model.add(words, vectors)

if you want you can then save it

model.save('mymodel')

and later just load it

model = KeyedVectors.load('mymodel')