This repo is part of a system which collects temperature data, trains a model, and displays the results on a Flask website.
The end-to-end process is as follows:
- Temperature and humidity readings are recorded from Raspberry Pi, and uploaded to DynamoDB using a bash script run every 10 minutes from the Pi
- An LSTM model is trained (previously with notebooks, in the future with an automated pipeline) using historic readings, and the model is converted to tensorflow-lite format
- The model is deployed on the flask website.
- A web front end shows recent readings and uses the trained ML model to predict future values, hosted on AWS Elastic Beanstalk.
This repo contains the code for parts 1 and 2. Parts 3 and 4 is stored in a different repo to separate the front-end from the training process, and to allow the web application to run on a smaller instance that does not also need to perform model training.
Run poetry run python src/main.py
Testing: poetry run pytest (only a few added so far)
- Currently exploring setting up a serverless endpoint, following this notebook, but unclear whether this is possible for already existing pretrained models. So far the endpoint has not accepted the pretrained LSTM.
Next steps
- Set up Dockerfile or S3 to automate model retraining and save model, scaler, and training data to S3
- Continue exploring serverless endpoint, if this cannot be done for tensorflow then try an AWS Lambda