Final project for Le Wagon Data Science bootcamp.
The repository now ships with a ready-to-use Streamlit front-end located in
streamlit_app.py. The entry point relies on a couple of
optional assets – a trained TensorFlow model for the image classifier and a CSV
dataset for the recipe search index. The interface surfaces clear error
messages when those assets are missing so you can supply them in the environment
that suits your deployment best.
Provide the following environment variables (or secrets.toml entries) to
point the application towards your assets:
| Variable | Description |
|---|---|
COOK_MODEL_PATH |
Path to the TensorFlow .h5 weights file. Defaults to notebooks/model.h5. |
COOK_LABELS_PATH |
Optional JSON file containing the class labels. Can be a list (["pizza", "salad"]) or a dictionary mapping indices to names. |
COOK_TRAIN_IMAGE_DIR |
Folder containing the training images (used to infer label names when COOK_LABELS_PATH is not supplied). |
COOK_DATASET_PATH |
CSV file with the recipe dataset. Defaults to project_cook/data/full_dataset.csv. |
COOK_INDEX_DIR |
Directory where the Whoosh search index should be stored. Defaults to new_index. |
Only the model path is required to unlock image predictions. Recipe search works when a dataset is provided; the index is created automatically on first use.
Install the dependencies and start Streamlit:
pip install -r requirements.txt
streamlit run streamlit_app.pyWhen deploying to Streamlit Cloud make sure the above environment variables are configured in the project settings.