機器學習的開源工具包：Waffles

jopen 11年前發布 | 22K 次閱讀 Waffles 機器學習

Waffles英文原意是蜂蜜甜餅（見logo），在這里卻指代一個非常強大的機器學習的開源工具包，基于C++語言開發。

Waffles里包含的算法特別多，涉及機器學習的方方面面，推薦系統位于其中的Waffles_recommend tool，大概只占整個Waffles的1/10的內容（其它還有分類、聚類、采樣、降維、數據可視化、音頻處理等許許多多工具包，估計能與之媲美的也就數Weka了）。

Waffles just works in more situations:
- All of our supervised learners support both classification and regression.
- All of them implicitly support multi-dimensional labels.
- Developing learning algorithms for Waffles is easier because:
  - It automatically imputes missing values if your algorithm is not robust to missing values.
  - It automatically discretizes data if your algorithm only supports discrete values.
  - It automatically converts nominal values to categorical distributions if your algorithm only supports continuous values.
  - When necessary, Waffles algorithms automatically normalize and denormalize data to fall within the ranges expected by particular algorithms.
  - It automatically converts predicted values back to their original form, so the user can freely operate on arbitrary data with any algorithm.
Waffles algorithms can automatically tune their own parameters.
Waffles is especially rich in dimensionality reduction techniques, but also provides numerous supervised learning algorithms, collaborative filtering algorithms, data transformations, linear-algebraic operations, optimization methods, data generation tools, visualization tools, etc.
Waffles tools are fast because they are developed as native applications.
Waffles does not depend on any platform or virtual machine, so it will not limit your distribution options.
Waffles is distributed under the permissive LGPL license.
Waffles apps are script-friendly, and are designed to be used with automated processes.
Waffles apps are thin wrappers around functionality in a well-documented C++ class library.