1. Feature extraction utilities

This package also features helpers to fetch larger datasets and parameters commonly used by the machine learning community to benchmark algorithm on data that comes from the ‘real world’.

1.1. Overfeat

sklearn-theano wraps a variety of trained neural networks for use as “black-box” transforms and classifiers. One of these networks in is known as Overfeat, seen in the publication:

  1. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, Y. LeCun. OverFeat: Integrated Recognition, Localization, and Detection using Convolutional Networks, International Conference on Learning Representations (ICLR 2014), April 2014.

Various layers of this neural network can be used for transformation, classification, and localization.

../images/plot_multiple_localization_0011.png

Examples:

  • plot_multiple_localization.py

1.2. GoogLeNet

Another state of the art neural network from Google is known as GoogLeNet:

  1. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, A. Rabinovich. Going Deeper with Convolutions, arXiv:1409.4842, September 2014.

Various layers of this neural network can be used for transformation and classification.

../images/plot_comparison_0011.png

Examples:

  • plot_comparison.py