Example of LSTM sequence inference with two plot graphs, one plotting the sequences, the other showcasing predicted and real values.
Author(s):
- Matej Volansky (2024)
- as a part of the team work preparation in 2023/2024
Docker image size: 1.26 GB
First train your model and save the .pt state dictionary after training (considering you're using PyTorch).
Update the model/model.py with your model. Don't forget that you have to create test data for the inference the same way you do for training, so update it in utils/sequence_dataset.py if you change the way you generate datasets.
Note: The technologies demostrated in this example are just selected ones. You can chose other technologies as you like (or the best you deal with).
The way sequences are plotted in this project depends on the way the data are sent between backend (app.py) and frontend. If you are using this example for some other sequences, change the way they are loaded and plotted, since this example is pretty specific.
If VanillaJS seems like a lot of trouble for plotting graphs for you, you can change the architecture to run with two docker images ( one for frontend, another for backend ), and in the frontend one have React run in it (for example, or any framework you prefer). Or you could even create the plots at the backend with python and then send them to frontend as images too.
To use this inference, just run
docker compose up
After successful build, your server will be available at http://localhost:8080/
For manual running without docker you have to create a python virtual environment.
python -m venv venv
source venv/bin/activate # on Linux based distros
source venv\Scripts\activate.ps1 # on Windows (Powershell)
source venv\Scripts\activate.bat # on Windows (cmd)
To install the PyTorch CPU version run
pip install torch==2.2.2+cpu torchvision==0.17.2+cpu -f https://download.pytorch.org/whl/torch_stable.html
After that go ahead and run
pip install -r requirements.txt