Siamese Deep Neural Networks for Semantic Text Similarity PyTorch
A repository containing comprehensive Neural Networks based PyTorch implementations for the semantic text similarity task, including architectures such as:
- Siamese LSTM
- Siamese BiLSTM with Attention
- Siamese Transformer
- Siamese BERT.
Usage
- install dependencies
pip install -r requirements.txt
- download spacy en model for tokenization
python -m spacy download en
Siamese LSTM
## init siamese lstm
siamese_lstm = SiameseLSTM(
batch_size=batch_size,
output_size=output_size,
hidden_size=hidden_size,
vocab_size=vocab_size,
embedding_size=embedding_size,
embedding_weights=embedding_weights,
lstm_layers=lstm_layers,
device=device,
)
## define optimizer
optimizer = torch.optim.Adam(params=siamese_lstm.parameters())
## train model
train_model(
model=siamese_lstm,
optimizer=optimizer,
dataloader=sick_dataloaders,
data=sick_data,
max_epochs=max_epochs,
config_dict={"device": device, "model_name": "siamese_lstm"},
)
Siamese BiLSTM with Attention
Siamese BiLSTM with Attention Example
## init siamese lstm
siamese_lstm_attention = SiameseBiLSTMAttention(
batch_size=batch_size,
output_size=output_size,
hidden_size=hidden_size,
vocab_size=vocab_size,
embedding_size=embedding_size,
embedding_weights=embedding_weights,
lstm_layers=lstm_layers,
self_attention_config=self_attention_config,
fc_hidden_size=fc_hidden_size,
device=device,
bidirectional=bidirectional,
)
## define optimizer
optimizer = torch.optim.Adam(params=siamese_lstm_attention.parameters())
## train model
train_model(
model=siamese_lstm_attention,
optimizer=optimizer,
dataloader=sick_dataloaders,
data=sick_data,
max_epochs=max_epochs,
config_dict={
"device": device,
"model_name": "siamese_lstm_attention",
"self_attention_config": self_attention_config,
},
)
Siamese Transformer
## init siamese bilstm with attention
siamese_transformer = SiameseTransformer(
batch_size=batch_size,
vocab_size=vocab_size,
embedding_size=embedding_size,
nhead=attention_heads,
hidden_size=hidden_size,
transformer_layers=transformer_layers,
embedding_weights=embedding_weights,
device=device,
dropout=dropout,
max_sequence_len=max_sequence_len,
)
## define optimizer
optimizer = torch.optim.Adam(params=siamese_transformer.parameters())
## train model
train_model(
model=siamese_transformer,
optimizer=optimizer,
dataloader=sick_dataloaders,
data=sick_data,
max_epochs=max_epochs,
config_dict={"device": device, "model_name": "siamese_transformer"},
)
Siamese BERT
from siamese_sts.siamese_net.siamese_bert import BertForSequenceClassification
## init siamese bert
siamese_bert = BertForSequenceClassification.from_pretrained(model_name)
## train model
trainer = transformers.Trainer(
model=siamese_bert,
args=transformers.TrainingArguments(
output_dir="./output",
overwrite_output_dir=True,
learning_rate=1e-5,
do_train=True,
num_train_epochs=num_epochs,
# Adjust batch size if this doesn't fit on the Colab GPU
per_device_train_batch_size=batch_size,
save_steps=3000,
),
train_dataset=sick_dataloader,
)
trainer.train()
89 Dec 18, 2022
912 Jan 8, 2023
224 Jan 4, 2023
48 Dec 18, 2022
12 Dec 18, 2022
15 Apr 16, 2022
1 May 24, 2022
1 Dec 19, 2021
![[CVPR 2022 Oral] Crafting Better Contrastive Views for Siamese Representation Learning](https://github.com/xyupeng/ContrastiveCrop/raw/main/figs/motivation.png)
249 Dec 28, 2022
892 Dec 28, 2022
9 Dec 27, 2022
32 Jan 2, 2023
159 Dec 22, 2022
28 Nov 25, 2022
101 Dec 29, 2022
203 Nov 30, 2022
432 Dec 17, 2022
72 Nov 9, 2022
834 Dec 30, 2022