yaplf (Yet Another Python Learning Framework) is an extensible machine learning framework written in python, including different models and learning algorithms for classification; currently, the framework supports:
perceptrons (either trained through Rosenblatt or gradient descent algorithm),
multilayer perceptrons with generic number of hidden layers (trained through error backpropagation algorithm with customizable learning rate, momentum, error threshold, and arbitrary differentiable activation functions on the various layers) ,
SV classifiers (both in the original and soft-margin formulation, and providing an extensible set of kernel functions),
an implementation of special SV classification algorithms for data of variable quality.
All models and algorithms are accessed in a consistent way, so that it is easy to compare the performances of different learning approaches on a same data set.
The system is opened to the addition of new models and algorithms for classification, as well as to other machine learning techniques such as regression or clustering. Moreover, it provides model-independent support for k-fold cross validation, extensive support for generating 2D and 3D images exploiting different graphic libraries (matplotlib and sage are currently supported and automatically detected), and basic support for multi-threading.
yaplf can be used out of the box both within a plain python environment or coupled with the open source sage system, either in form of extended python scripts or within a web-based interactive notebook.
Finally, the system includes the support for graphic observation of iterative learning algorithms, including:
the visualization of error versus iteration number (with arbitrary error metrics), and
the visualization of trajectories in the model space.
These features make yaplf an interesting tool also for teaching machine learning.
- Changes to previous version:
Initial Announcement on mloss.org.
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.