[英]How to convert embedding matrix to torch.Tensor
I am new to pytorch and not sure how to convert an embedding matrix to a torch.Tensor
type我是 pytorch 的新手,不确定如何将嵌入矩阵转换为torch.Tensor
类型
I have 240 rows of input text data that I convert to embedding using Sentence Transformer
library like below我有 240 行输入文本数据,我使用如下所示的Sentence Transformer
库将其转换为嵌入
embedding_model = SentenceTransformer('bert-base-nli-mean-tokens')
features = embedding_model.encode(df.features.values)
Now this features
is a numpy.ndarray
of shape (240, 768)
现在这个features
是一个numpy.ndarray
形状(240, 768)
I have defined the model as我已经将模型定义为
class NClassifier(nn.Module):
def __init__(self, input_dim, embedding_dim, hidden_dim, tagset_size):
super(NClassifier, self).__init__()
self.hidden_dim = hidden_dim
self.word_embeddings = nn.Embedding(input_dim, embedding_dim)
# The LSTM takes word embeddings as inputs, and outputs hidden states
# with dimensionality hidden_dim.
self.lstm = nn.LSTM(embedding_dim, hidden_dim)
# The linear layer that maps from hidden state space to code space (output clases)
self.hidden2code = nn.Linear(hidden_dim, tagset_size)
def forward(self, features):
embeds = self.word_embeddings(features)
lstm_out, _ = self.lstm(embeds.view(len(features), 1, -1))
code_space = self.hidden2code(lstm_out.view(len(features), -1))
code_scores = F.log_softmax(code_space, dim=1)
return code_scores
INPUT_DIM = 240
EMBEDDING_DIM = 768
HIDDEN_DIM = 256
OUTPUT_DIM = 34
model = NClassifier(INPUT_DIM, EMBEDDING_DIM, HIDDEN_DIM, OUTPUT_DIM)
Now when I do scores = model(features)
I get error as features
is NOT a tensor.现在,当我执行scores = model(features)
我收到错误,因为features
不是张量。 I see the example of converting the input to tensor here but it is not clear to me.我在这里看到了将输入转换为张量的示例,但我不清楚。
Can anyone please help?有人可以帮忙吗?
A numpy.ndarray
can be converted to a torch.Tensor
with the torch.tensor()
function, like this:甲numpy.ndarray
可以转换为一个torch.Tensor
与torch.tensor()
函数,如下所示:
features_tensor = torch.tensor(features)
Does that work for you?那对你有用吗?
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.