Comments (4)
Sorry,maybe I have not described this problem clearly.
For example,I have a label set {"label1":1,"label2":2},so the function size
returns 3 cause of the length of instance is 2.
When use_crf=False
,
# add two more label for downlayer lstm, use original label size for CRF
label_size = data.label_alphabet_size
data.label_alphabet_size += 2
# so the hidden2tag
self.hidden2tag = nn.Linear(data.HP_hidden_dim, data.label_alphabet_size)
the output size of hidden2tag is 5,
when use _, tag_seq = torch.max(outs, 1)
,the result maybe 0-4,but the label alphabe only has two labels,index 1 and index 2.
Thank you.
from ncrfpp.
I think I get your point. I also had this concern during writing this framework.
The "START" and "END" is for the CRF calculation. In the CRF layer, I set some of the default transition scores to -10000 to avoid the "START" and "END" output.
To keep the code simple, the "START" and "END" are also added in the model with softmax output. Theoretically, the model may decode some invalid labels "START/END" but it is almost impossible in real data. When the model is trained with the training data, it will not decode the invalid labels as they do not exist in the training data.
from ncrfpp.
Thank you.
In other words,once after training,the _, tag_seq = torch.max(outs, 1)
almost will not produce invalid output.
from ncrfpp.
Exactly!
from ncrfpp.
Related Issues (20)
- Conll 使用内存大 HOT 2
- nbest score HOT 1
- Provision for Custom Features? HOT 1
- Bug in forward method when calling _viterbi_decode, mask is not provided
- about requirement HOT 1
- Can I just use the CRF layer? HOT 1
- 大神你好,请教一些关于报错的问题 HOT 1
- About "tcmalloc: large alloc" message and training aborted HOT 3
- Difference between main_parse.py and main.py HOT 1
- Different results on CPU versus GPU HOT 1
- Test Data Format for Chunking
- Experimenting with Transformer models? HOT 1
- Decode config parameter when not using CRF HOT 1
- Reading fasttext word embeddings HOT 2
- About MAX_SENTENCE_LENGTH parameter HOT 2
- Decode error with nbest=0 when not using CRF
- Resume model training from a checkpoint
- 训练一段时间后,f1为-1 HOT 2
- Unknown labels in decoding
- Support trained model serialization to ONNX format HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from ncrfpp.