ragavvenkatesan/yann

This toolbox is support material for the book on CNN (http://www.convolution.network).

yann: version 1.0rc1

• An even easier, simpler and versatile CNN tool alternative to Lasagne and Keras.
• Promotes easy learnability and practice of CNNs using theano.
• Modular and plug and play modules and layers.
• Implementations of most commonly used techniques and popular papers in CNN research.

Originally written for personal research, this toolbox will be most useful for those who are interested in starting deep CNNs to make a fast transition into theano and developing CNNs. It was called samosa back then. Now yann has evolved into this toolbox for CNNs. Yann is designed as a supplement to an upcoming book on Convolutional Neural Networks and also the toolbox of choice for a graduatelevel deep learning course under developement at CIDSE ASU.

Because of this reason, Yann is specifically designed to be intuitive and easy to use for beginners. That does not compromise Yann of any of its features. It is still a good choice for a toolbox for running pre-trained models and build complicated, non-vannilla CNN architectures that are not easy to build with the other toolboxes.This will also be useful for industry when teting out prototypes.

There is a documentation page that has some details on how to run the code and how to use it. For a quick start into the toolbox, you can use this guide.

Thanks for using the code, hope you had fun. Ragav Venkatesan

http://www.ragav.net

yann's Issues

Name 'Preprocessing' is not defined in /yann/utils/datasets.py

While embarking through your quick-start documentation for YANN I encountered an issue with datasets.py when cooking the MNIST data from Skdata.

Attached is a screenshot featuring a python interpreter and the associated behavior of the cook_mnist() function included in datasets.py where the name 'preprocessing' appears undefined:

To note - the 'preprocessing' name is mentioned in a comment at the beginning of datasets.py

NameError: name 'cost' is not defined in /yann/network.py

While working through the quick-start documentation to gain more YANN knowledge for better test design I encountered an error with the function net.test()

I have a pull request ready to be created that will also fix misc. import issues that allow for net.cook() to work fully before calling net.train()

Below is a screenshot of the python interpreter trace and the source of the issue within network.py

Need to redo readme and gh-pages.

Priority is to have detailed instructions on

• Pr-requisites and their setup.
• How to setup the code
• How to set up various architectures.
• How to run.

Second Priority is to come up with full dev level documentation using Sphinx.