Release 0.7: Native Layers, Batch Normalization, pandas.DataFrame and Memory Mapped Arrays

The sixth official release of scikit-neuralnetwork — version 0.7 — 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

This release includes a variety of improvements such as native layers (that let you include any Lasagne code directly), batch normalization, support for various data formats such as pandas.DataFrame and numpy Memory Mapped Arrays, as well as a variety of fixes and improvements. Read on for details!

Consult the stable documentation for more information:
http://scikit-neuralnetwork.readthedocs.org/en/stable/

The release file is attached here for reference too.

Features

• Native Layers to support backend features directly. #195
• Back to official Theano releases, now that Theano-0.8.0 is out. #191
• Batch normalization support, specified per-layer. #187
• Support for pandas.DataFrame as input data to fit, both X and y. #186
• Multi-feature classification returns list of arrays for convenience. #185
• Performance improvements and fixes in batch iterator. #170
• Support for numpy's Memory Mapped Arrays. [39b40d8]

Bug Fixes

• Calling set_parameters() even before initialize works. [0d42fa7]
• Fix unused parameter warning when using convolution layers. [1f6fb25]
• Fixed regularization support, L1 and L2 were not working correctly. [74e1778]
• Chaining get_parameters and set_parameters on another network. [24d9b3c]

Release 0.6: Bug Fixes, Multithreading on CPU, Voting Ensemble Support

The sixth official release of scikit-neuralnetwork — version 0.6 — 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

This release includes a large number of bug fixes (e.g. unpickled binary classifiers), small features (leveraging voting ensembles from sklearn) and adds a helper function for running on multiple CPU threads. 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.

Features

• Setting the number of CPU threads via platform module. #165
• Support and tests for use with Voting Ensemble. #154

Bug Fixes

• Fix for datasets strictly smaller than the batch size. #158
• Binary classifier works correctly after unpickling. #161

Release 0.5: Callbacks, Weighted Samples, Output Convolution, Exponential Linear Activation

The 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]

Release 0.4: Lasagne Backend, Deprecating PyLearn2

The fourth official release of scikit-neuralnetwork — version 0.4 — 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 mainly features a new backend for Lasagne, which needs to be installed separately. The backend for PyLearn2 is still the default but will be removed in the next release.

Consult the documentation for more information:
http://scikit-neuralnetwork.readthedocs.org/en/stable/

The release file is attached here for reference too.

Release 0.3: Multiple Backend Support, Bug Fixes

The third official release of scikit-neuralnetwork — version 0.3 — 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

This release mainly features a refactor to support multiple backend neural network implementations, and a variety of minor fixes for pylearn2. In particular: #64, #66, #67, #71, #72, #73, #75, #78, #83, #85, #86, #89, #90.

Consult the documentation for more information:
http://scikit-neuralnetwork.readthedocs.org/en/stable/

The release file is attached here for reference too.

Release 0.2: Auto-Encoders, Performance, Regularization

The second official release of scikit-neuralnetwork — version 0.2 — 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

This release mainly features a new auto-encoder module that can act as a scikit-learn transform, and used as unsupervised pre-training. In particular: #43, #47, #48, #49, #52, #53, #55, #58, #61, #62.

Consult the documentation for more information:
http://scikit-neuralnetwork.readthedocs.org/en/stable/

The release file is attached here for reference too.

Release 0.1: Multi-Layer Perceptrons

The first official release of scikit-neuralnetwork — version 0.1 — 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

It features a large range of features from different activation types, to layer types, to learning rules and dataset types. Consult the documentation for more details:
http://scikit-neuralnetwork.readthedocs.org/en/latest/

The release file is attached here for reference too.