Cinnamon Examples

This project contains a complete example of a machine-learning pipeline.

In particular, we'll see how

• Define a DataLoader
• Define two data Processor, one for tf-idf features and one for one-hot label encoding
• Define a SVM classifier Model
• Train and evaluate our classifier with a Benchmark

We consider the IMDB data to perform a binary classification for sentiment analysis.

Documentation

Check the online documentation of cinnamon-examples for a detailed walkthrough on the project.

Usage

Want to use the Component and Configuration of this example project?

You can quickly integrate them into your project by cloning this Github repo and add it to Registry.setup().

Registry.setup(directory=...,
               external_directories=[
                   "path/where/cinnamon_examples/is/stored",
               ])

Done! Cinnamon will automatically import all registered Configuration under configurations folders.

Note: Remember to first install all requirements to allow successful code import.

We then can define any registration that uses Component or Configuration declared in this repo.

For instance, let's say we want to define a new model configuration:

    @register(name='model', tags={'custom'}, namespace='my_own')
    def register_custom_model_config():
        config = SVCModelConfig.default()
        
        config.C = 0.5
        config.kernel = 'rbf'
        
        return config

Or that we want to re-use IMBDLoader with our custom benchmark.

    class CustomBenchmarkConfig(Configuration):
    
        @classmethod
        @register_method(name='benchmark', tags={'custom'}, namespace='my_own')
        def default(cls):
            config = super(cls).default()
            
            config.add(name='data_loader',
                       value=RegistrationKey(name='data_loader',
                                         tags={'imdb'},
                                         namespace='examples'))
            
            config.add(name='model',
                       value=RegistrationKey(name='model',
                                             tags={'imdb', 'lstm'},
                                             namespace='my_own'))
            
            ...
        
            return config

Contact

Don't hesitate to contact:

• Federico Ruggeri @ federico.ruggeri6@unibo.it

for questions/doubts/issues!