Apprentissage paresseux
L'apprentissage paresseux est un type de apprentissage automatique where the model does not attempt to generalize from the données d'entraînement until a query is made. This approach contrasts with eager learning algorithms, which build a model during the training phase and make predictions based on the generalized model.
In lazy learning, the system stores the training data and waits until a request for a prediction is received. When a prediction is needed, the algorithm uses the stored instances to make a decision. This method is particularly useful in scenarios where the data is complex and diverse, potentially leading to better predictions without the bias of an oversimplified model.
Quelques exemples courants d'algorithmes d'apprentissage paresseux incluent :
- k-Plus proches voisins (k-NN) : This algorithm classifies a new instance based on the majority class of its ‘k’ nearest training instances in the feature space.
- Raisonnement basé sur des cas (CBR) : This approach solves new problems based on the solutions of similar past problems.
Lazy learning has its advantages and disadvantages. One major advantage is that it can be more flexible and can adapt to nouvelles données without needing to retrain a model. However, it can also be computationally expensive at the time of prediction, especially if the dataset is large, as it requires the algorithm to consider all stored instances to make accurate predictions.
Dans l'ensemble, l'apprentissage paresseux est une machine puissante stratégie d'apprentissage that prioritizes immediate data retrieval and analysis over upfront model construction, making it suitable for specific applications where data patterns may frequently change.