GPU环境下spaCy+Numpy求相似词向量遇类型错误的解决咨询

在GPU环境中运行spaCy查找相似词向量的代码时，触发了TypeError，错误提示为：

Implicit conversion to a NumPy array is not allowed. Please use .get() to construct a NumPy array explicitly

原代码 #

import numpy as np
your_word = "country"
ms = nlp.vocab.vectors.most_similar(
    np.asarray([nlp.vocab.vectors[nlp.vocab.strings[your_word]]]), 
    n=10, )
words = [nlp.vocab.strings[w] for w in ms[0][0]]
distances = ms[2]
print(words)

报错栈 #

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[139], line 6
      1 import numpy as np
      3 your_word = "country"
      5 ms = nlp.vocab.vectors.most_similar(
----> 6     np.asarray([nlp.vocab.vectors[nlp.vocab.strings[your_word]]]), 
      7     n=10,
      8 )
     10 words = [nlp.vocab.strings[w] for w in ms[0][0]]
     11 distances = ms[2]

File cupy/_core/core.pyx:1475, in cupy._core.core._ndarray_base.__array__()

TypeError: Implicit conversion to a NumPy array is not allowed. Please use `.get()` to construct a NumPy array explicitly.

问题原因 #

GPU环境下，spaCy的词向量存储在CuPy数组中，直接用np.asarray()尝试将其转为NumPy数组会触发隐式转换限制，CuPy不允许这种隐式操作，必须显式调用*.get()*方法将CuPy数组转为NumPy数组。

修复方案 #

方案1：显式转为NumPy数组（CPU运算） #

import numpy as np
your_word = "country"
# 显式用.get()将CuPy数组转为NumPy数组
word_vector = nlp.vocab.vectors[nlp.vocab.strings[your_word]].get()
ms = nlp.vocab.vectors.most_similar(
    np.asarray([word_vector]), 
    n=10, )
words = [nlp.vocab.strings[w] for w in ms[0][0]]
distances = ms[2]
print(words)

方案2：直接使用CuPy数组（GPU加速，推荐） #

避免数据回传CPU，保持GPU运算效率：

your_word = "country"
word_vector = nlp.vocab.vectors[nlp.vocab.strings[your_word]]
# 将向量转为(1, 维度)的形状，满足most_similar的输入要求
ms = nlp.vocab.vectors.most_similar(
    word_vector.reshape(1, -1), 
    n=10, )
words = [nlp.vocab.strings[w] for w in ms[0][0]]
distances = ms[2]
print(words)

内容的提问来源于stack exchange，提问作者onder