Release 0.5: Callbacks, Weighted Samples, Output Convolution, Exponential Linear Activation
Pre-releaseThe fifth official release of scikit-neuralnetwork
— version 0.5 — is available on PYPI from the following URL:
https://pypi.python.org/pypi/scikit-neuralnetwork
Or simply type this to install the latest version directly from the command-line with pip
:
pip install scikit-neuralnetwork lasagne
This release removes the PyLearn2 backend and makes Lasagne default. It also includes many new features and improvements. Read on for details!
Consult the documentation for more information:
http://scikit-neuralnetwork.readthedocs.org/en/stable/
The release file is attached here for reference too.
Major Features
Dataset masking, aka. sample weighting. #135
Generic callback implementation. #133
Output convolution layers. #137
Exponential linear units. #138
Upscaling in convolution. [81e9a46]
Improvements & Fixes
Warning if no iterations specified. [58fcb3c]
Saving best network automatically. [2694667]
Easy access to network parameters. [42397ef]
Set parameters on initialized network. [ada168b]
Support for classes_
property. [fd1987e]
Correct validation cost display. [95b3b9b]
Training progress bar display. [1b46f2b]
Stability check on training data if no validation. [3a06089]