使用词嵌入进行分类

Question

我正在尝试使用词嵌入进行分类，但我遇到了typeError问题。

# glove word embeddings
import numpy as np  
embeddings_index = {}
with open('glove.6B/glove.6B.50d.txt', 'r') as f:
    for line in f:
        values = line.split()
        word = values[0]
        coefs = np.asarray(values[1:], dtype='float32')
        embeddings_index[word] = coefs```

# transform text (a title) to an embedding by averaging word embeddings
def get_mean_embeddings(docs,embeddings):
    means = []
    dim = len(embeddings.values()[0])
    for doc in docs :
        words = tokenizer.tokenize(doc)
        means.append(np.mean([embeddings[w] if w in embeddings else np.zeros(dim) for w in words], axis=0)) 
    return np.array(means)```

def get_mean_embeddings(docs,embeddings):
    dim = len(embeddings.values()[0])
    return np.array([
                np.mean([embeddings[w]
                         for w in tokenizer.tokenize(doc) if w in embeddings] or
                        [np.zeros(dim)], axis=0)
                for doc in docs
            ])

import sklearn.svm as svm
from sklearn.metrics import f1_score
clf = svm.SVC(kernel='rbf')
f1_scores = []
for g in genres:
    genre_data = balanced_data[g]
    train,test = train_test_split(genre_data,train_size = 0.6)
    train_feature_matrix = get_mean_embeddings(train['title'],embeddings)
    test_feature_matrix = get_mean_embeddings(test['title'],embeddings)
    clf.fit(train_feature_matrix,train[g])
    y_pred = clf.predict(test_feature_matrix)
    f1_scores.append(f1_score(test[g],y_pred))
    print('for "%s" , f1 score = %.2f' %(g,f1_scores[-1]))

print ('average f1 score over all genres : %.2f ' %(np.mean(f1_scores)))

预期和实际结果：

for "sci-fi" , f1 score = 0.70
for "horror" , f1 score = 0.68
for "fantasy" , f1 score = 0.62
for "adventure" , f1 score = 0.66
for "thriller" , f1 score = 0.63
for "mystery" , f1 score = 0.58
for "romance" , f1 score = 0.62
for "crime" , f1 score = 0.56
for "drama" , f1 score = 0.59
for "action" , f1 score = 0.67
for "comedy" , f1 score = 0.62
for "documentary" , f1 score = 0.64
for "war" , f1 score = 0.65
average f1 score over all genres : 0.63

错误：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-33-7c91ab021935> in <module>
      6     genre_data = balanced_data[g]
      7     train,test = train_test_split(genre_data,train_size = 0.6)
----> 8     train_feature_matrix = get_mean_embeddings(train['title'],embeddings)
      9     test_feature_matrix = get_mean_embeddings(test['title'],embeddings)
     10     clf.fit(train_feature_matrix,train[g])

<ipython-input-25-0a52cf917522> in get_mean_embeddings(docs, embeddings)
      1 def get_mean_embeddings(docs,embeddings):
----> 2     dim = len(embeddings.values()[0])
      3     return np.array([
      4                 np.mean([embeddings[w]
      5                          for w in tokenizer.tokenize(doc) if w in embeddings] or

TypeError: 'dict_values' object is not subscriptable

Answer 1

问题是，in Python 3, dict_values is merely a view and not a list。

如果你想得到第一个元素的长度，你必须替换

dim = len(embeddings.values())

通过以下方式：

列表dim =len(

(embeddings.values()

有关更多信息，请参阅：Python: how to convert a dictionary into a subscriptable array?