Comments (3)
Wait, once I load the model, I can see the embedding layer already fine tuned. This mapping of stoi and itos while take place from this layer right? Correct me if I'm wrong.
from pytorch-sentiment-analysis.
You don't need to re-load the word embedding parameters with model.embedding.weight.data.copy_(TEXT.vocab.vectors)
as they will already be saved with the model when you do torch.save
.
The only issue is you need to make sure the vocabulary of your new data is exactly the same as the vocabulary of your old data.
Say your original vocab stoi was {'the': 0, 'cat': 1, 'sat': 2}
and your vocab stoi for the new data was {'the': 0, 'bird': 1, 'flew': 2}
. Using your new data/vocab, if you index the sentence "the bird flew"
you will get a tensor of [0, 1, 2]
. However, when you pass this tensor through the embedding layer, you will actually get the embeddings for the sentence "the cat sat"
.
One way to get around this is by finishing up your training on the first set of data by turning the vocab and corresponding tensors into a dictionary, i.e. {'the': [0.2, 0.1, 0.3], 'cat': [0.3, 0.6, -0.2], ...
and then saving it (as a .json file or similar). Then, when using your new data, you'll create your vocab without using pre-trained vectors, and then create a function that loads the saved vocab-tensor dictionary and iterates over all the keys. If you find a key in the loaded vocab that is also in your current vocab you can initialize that column of the embedding weight tensor (indicated by the stoi value) with your loaded vocab-tensor value.
Let me know if that isn't clear and make a quick gist.
from pytorch-sentiment-analysis.
Yes, I understood. I was loading the same train set into testing and saw the mismatch of stoi/itos. That makes perfect sense. Thanks a lot for making it clear.
I have another question about one-hot encoded targets. Ill generate a new issue.
from pytorch-sentiment-analysis.
Related Issues (20)
- The train_data built from my own dataset after following the Appendix A looks wrong HOT 1
- migrating to the new API HOT 4
- .squeeze(1) HOT 5
- for word embedding in RNN model HOT 2
- Representation of similar words HOT 1
- Using a target size (torch.Size([64, 1])) that is different to the input size (torch.Size([304800, 1])) is deprecated. Please ensure they have the same size HOT 7
- train_test_split in LSTM HOT 2
- pad sequence in my dataset
- 6 - Transformers for Sentiment Analysis HOT 1
- Question in fasttext HOT 1
- ModuleNotFoundError
- How can I predict on one example?
- User Interface
- Got error" 'lengths' argument should be a 1D CPU int64 tensor, but got 1D cuda:0 Long tensor " HOT 10
- where is the trained model parameters ? HOT 1
- TypeError HOT 2
- Multi-class Sentiment Analysis: How to use custom dataset?
- how did you build torchtext from source HOT 4
- how does pytorch pad sentences ? HOT 3
- Pad Sequence error HOT 2
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 pytorch-sentiment-analysis.