Worked pipeline

This section walks through a complete machine-learning pipeline, end to end.

It performs binary sentiment analysis on the IMDB dataset using a Support Vector Machine. Each stage — data loading, preprocessing, modelling, and evaluation — is a separate component and configuration, wired together by the Registry.

        flowchart LR
    L["IMDBLoader<br/>data_loader[imdb]"]
    T["TfIdfProcessor<br/>processor[tf-idf]"]
    E["LabelProcessor<br/>processor[label]"]
    M["SVCModel<br/>model[svc]"]
    B["SVCBenchmark<br/>benchmark[svc]"]
    B --> L
    B --> T
    B --> E
    B --> M
    

The benchmark holds the four keys and builds each child when run() needs it. Every arrow is a dependency field, so cmn-check reports a typo in any of them before anything is loaded.

pip install -e ".[examples]"
python -m examples.demos.demo_benchmark

Note

The dataset is downloaded on first run.

If you have not met the concepts yet, start with the Tutorial instead: seven short runnable files that build them up one at a time, needing nothing beyond cinnamon itself. This section assumes them.