Pipeline 101

pipeline.py

from pyannote.pipeline import Pipeline
import chocolate

class MyPipeline(Pipeline):
    def __int__(self):
        super().__init__()
        self.damping = chocolate.uniform(...)
        self.preference = chocolate.uniform(...)
    
    def __call__(self, item):
        # item = (X, y_true)
        X = item[0]
        y_pred = clustering(X, self.damping, self.preference)
        return y_pred
    
    def loss(self, item, y_pred)
        y_true = item[1]
        return f(y_true, y_pred)

from pyannote.pipeline import Optimizer
optimizer = Optimizer(MyPipeline(), '/tmp/toto.db')

result = optimizer.tune([(X1, y1), (X2, y2), (X3, y3)], n_iterations=100)

print(result['best']['loss'])


        
        

You can also use pyannote.pipeline.blocks.clustering.AffinityPropagationClustering as a "sub-pipeline".
Its hyper-parameters will automatically be added to the main pipeline:

...

def __init__(self):
    super().__init__()
    self.clustering = AffinityPropagationClustering()

def __call__(self, item):
    X, y_true = item
    return self.clustering(X)   # its parameters are already set by with_params

...