Algorithm support in Mahout
The implementation of algorithms in Mahout can be categorized into two groups:
- Sequential algorithms: These algorithms are executed sequentially and so cannot use Hadoop's scalable processing. These algorithms are usually the ones derived from Taste (this was a separate project. It was a non Hadoop based recommendation engine).
Examples of these algorithms are user-based collaborative filtering, logistic regression, Hidden Markov Model, multi-layer perceptron, and singular value decomposition.
- Parallel algorithms: These algorithms can support petabytes of data using Hadoop's map reduce parallel processing.
Examples of these algorithms are Random Forest, Naïve Bayes, Canopy clustering, K-means clustering, spectral clustering, ...