Adi*_*yal 7 python cosine-similarity word-embedding
我正在尝试使用CLIP来计算字符串之间的相似度。（我知道 CLIP 通常用于文本和图像，但它也应该只适用于字符串。）
我提供了简单文本提示的列表，并计算它们嵌入之间的相似度。相似之处消失了，但我不知道我做错了什么。
import torch
import clip
from torch.nn import CosineSimilarity
cos = CosineSimilarity(dim=1, eps=1e-6)
def gen_features(model, text):
tokens = clip.tokenize([text]).to(device)
text_features = model.encode_text(tokens)
return text_features
def dist(v1, v2):
#return torch.dist(normalize(v1), normalize(v2)) # euclidean distance
#return cos(normalize(v1), normalize(v2)).item() # cosine similarity
similarity = (normalize(v1) @ normalize(v2).T)
return similarity.item()
device = "cuda" if torch.cuda.is_available() else "cpu"
model_name = "ViT-B/32"
model, _ = clip.load(model_name, device=device)
sentences = ["A cat", "A dog", "A labrador", "A poodle", "A wolf", "A lion", "A house"]
with torch.no_grad():
embeddings = [(sentence, gen_features(model, sentence)) for sentence in sentences]
for label1, embedding1 in embeddings:
for label2, embedding2 in embeddings:
print(f"{label1} -> {label2}: {dist(embedding1, embedding2)}")
Output
A cat -> A cat: 0.9999998211860657
A cat -> A dog: 0.9361147880554199
A cat -> A labrador: 0.8170720934867859
A cat -> A poodle: 0.8438302278518677
A cat -> A wolf: 0.9086413979530334
A cat -> A lion: 0.8914517164230347
A cat -> A house: 0.8724125027656555
A dog -> A cat: 0.9361147880554199
A dog -> A dog: 1.0000004768371582
A dog -> A labrador: 0.8481228351593018
A dog -> A poodle: 0.9010260105133057
A dog -> A wolf: 0.9260395169258118
A dog -> A lion: 0.886112630367279
A dog -> A house: 0.8852840662002563
A labrador -> A cat: 0.8170720934867859
A labrador -> A dog: 0.8481228351593018
A labrador -> A labrador: 1.000000238418579
A labrador -> A poodle: 0.7722526788711548
A labrador -> A wolf: 0.8111101984977722
A labrador -> A lion: 0.783727765083313
A labrador -> A house: 0.7569846510887146
A poodle -> A cat: 0.8438302278518677
A poodle -> A dog: 0.9010260105133057
A poodle -> A labrador: 0.7722526788711548
A poodle -> A poodle: 0.999999463558197
A poodle -> A wolf: 0.8539597988128662
A poodle -> A lion: 0.8460092544555664
A poodle -> A house: 0.8119628429412842
A wolf -> A cat: 0.9086413979530334
A wolf -> A dog: 0.9260395169258118
A wolf -> A labrador: 0.8111101984977722
A wolf -> A poodle: 0.8539597988128662
A wolf -> A wolf: 1.000000238418579
A wolf -> A lion: 0.9043934941291809
A wolf -> A house: 0.860664427280426
A lion -> A cat: 0.8914517164230347
A lion -> A dog: 0.886112630367279
A lion -> A labrador: 0.783727765083313
A lion -> A poodle: 0.8460092544555664
A lion -> A wolf: 0.9043934941291809
A lion -> A lion: 1.0000004768371582
A lion -> A house: 0.8402873873710632
A house -> A cat: 0.8724125027656555
A house -> A dog: 0.8852840662002563
A house -> A labrador: 0.7569846510887146
A house -> A poodle: 0.8119628429412842
A house -> A wolf: 0.860664427280426
A house -> A lion: 0.8402873873710632
A house -> A house: 0.9999997615814209
The results show that a dog is closer to a house than it is for a labrador 0.885 vs 0.848 which doesn't make sense. I've tried cosine similarity and euclidean distance to check whether the distance measure was wrong, but the results are similar. Where am I going wrong?
小智 7
如果您使用 ([提示数, 77, 512]) 输出中的文本嵌入CLIPTextModel，将它们展平 ([提示数, 39424]) 并应用余弦相似度，您将获得改进的结果。
此代码可让您测试这两个解决方案（[1,512] 和 [77,512]）。我正在 Google Colab 上运行它。
!pip install -U torch transformers
import torch
from torch.nn import CosineSimilarity
from transformers import CLIPTokenizer, CLIPModel, CLIPTextModel
cossim = CosineSimilarity(dim=0, eps=1e-6)
def dist(v1, v2):
return cossim(v1, v2)
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
models = [
'openai/clip-vit-base-patch16',
'openai/clip-vit-base-patch32',
'openai/clip-vit-large-patch14',
]
model_id = models[1]
tokenizer = CLIPTokenizer.from_pretrained(model_id)
text_encoder = CLIPTextModel.from_pretrained(model_id).to(torch_device)
model = CLIPModel.from_pretrained(model_id).to(torch_device)
prompts = [
"A cat", "A dog", "A labrador", "A poodle", "A wolf", "A lion", "A house",
]
text_inputs = tokenizer(
prompts,
padding="max_length",
return_tensors="pt",
).to(torch_device)
text_features = model.get_text_features(**text_inputs)
text_embeddings = torch.flatten(text_encoder(text_inputs.input_ids.to(torch_device))['last_hidden_state'],1,-1)
print("\n\nusing text_features")
for i1, label1 in enumerate(prompts):
for i2, label2 in enumerate(prompts):
if (i2>=i1):
print(f"{label1} <-> {label2} = {dist(text_features[i1], text_features[i2]):.4f}")
print("\n\nusing text_embeddings")
for i1, label1 in enumerate(prompts):
for i2, label2 in enumerate(prompts):
if (i2>=i1):
print(f"{label1} <-> {label2} = {dist(text_embeddings[i1], text_embeddings[i2]):.4f}")
您将获得 [1,512] 嵌入的相同值
A cat <-> A cat = 1.0000
A cat <-> A dog = 0.9361
A cat <-> A labrador = 0.8171
A cat <-> A poodle = 0.8438
A cat <-> A wolf = 0.9086
A cat <-> A lion = 0.8915
A cat <-> A house = 0.8724
A dog <-> A dog = 1.0000
**A dog <-> A labrador = 0.8481**
A dog <-> A poodle = 0.9010
A dog <-> A wolf = 0.9260
A dog <-> A lion = 0.8861
**A dog <-> A house = 0.8853**
A labrador <-> A labrador = 1.0000
A labrador <-> A poodle = 0.7723
A labrador <-> A wolf = 0.8111
A labrador <-> A lion = 0.7837
A labrador <-> A house = 0.7570
A poodle <-> A poodle = 1.0000
A poodle <-> A wolf = 0.8540
A poodle <-> A lion = 0.8460
A poodle <-> A house = 0.8120
A wolf <-> A wolf = 1.0000
A wolf <-> A lion = 0.9044
A wolf <-> A house = 0.8607
A lion <-> A lion = 1.0000
A lion <-> A house = 0.8403
A house <-> A house = 1.0000
但随着 [1,77,512] 嵌入，结果有所改善，现在狗离拉布拉多犬比离房子更近。尽管如此，您还是会得到有趣的结果，例如猫更像房子而不是贵宾犬。
A cat <-> A cat = 1.0000
A cat <-> A dog = 0.8880
A cat <-> A labrador = 0.8057
A cat <-> A poodle = 0.7579
A cat <-> A wolf = 0.8558
A cat <-> A lion = 0.8358
A cat <-> A house = 0.8024
A dog <-> A dog = 1.0000
**A dog <-> A labrador = 0.8794**
A dog <-> A poodle = 0.8583
A dog <-> A wolf = 0.8888
A dog <-> A lion = 0.8265
**A dog <-> A house = 0.8294**
A labrador <-> A labrador = 1.0000
A labrador <-> A poodle = 0.8006
A labrador <-> A wolf = 0.8182
A labrador <-> A lion = 0.7958
A labrador <-> A house = 0.7608
A poodle <-> A poodle = 1.0000
A poodle <-> A wolf = 0.7928
A poodle <-> A lion = 0.7735
A poodle <-> A house = 0.7623
A wolf <-> A wolf = 1.0000
A wolf <-> A lion = 0.8496
A wolf <-> A house = 0.8063
A lion <-> A lion = 1.0000
A lion <-> A house = 0.7671
A house <-> A house = 1.0000