#Deep Learning With Python

This repository is a Python implementation version of UFLDL(Unsupervised Feature Learning and Deep Learning) tutorial exercises, the codes are passed the test and get the same results as exepected.

tutorial homepage: http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial

IMPORTANT NOTES:

1. The datasets used in this repository can be found in UFLDL homepage.

2. I implememented softmax with bias, while the tutorial did not. I try softmax with bias and no bias, the result shows softmax with bias can achieved an exepected accuracy of 92.6% on MNIST dataset.

3. Scipy 0.17 will lead to python kenerl died when load the *.mat format file. It is a inherent bug in scipy 0.17, so I do not recommend you to use scipy 0.17 to run the codes.

##Prerequisites

• python 2.7
• numpy
• scipy
• or just Anaconda (strongly recommend)

##Exercises' core source code file are listed as follows:

Sparse Autoencoder

• sparseAutoencoder.py
• sparseAutoencoderTest.py

Preprocessing: PCA and Whitening

• whitening.py

Softmax Regression

• softmax.py

Self-Taught Learning and Unsupervised Feature Learning

• self-taughtLearningTest.py

Building Deep Networks for Classification(Stacked Sparse Autoencoder)

• stackedAutoencoder.py
• stackedAutoencoderTest.py

Linear Decoders with Autoencoders

• sparseAutoencoder.py
• SAEWithLinearDecoderTest.py

Working with Large Images(Convolutional Neural Networks)

• cnn.py
• cnnTest.py

---

If you have any questions, please feel free to contact with me. (guanghuitu@gmail.com or guanghuitu@foxmail.com)

Enjoy it!