Convolutional Neural Network Playground

• The aim of this repo is for a hands-on exploring and parcticing deep learning convolutional neural network upon various dataset.

Programming language

• Python 2.7 for master branch
• Python 3.5 for python35 branch

Library Dependencies

• master branch:
  • TensorFlow r0.11
  • numpy 1.11.2
  • scipy 0.18.1
  • scikit-learn 0.18.1
  • matplotlib 1.5.3
  • pandas 0.19.0
  • jupyter 1.0.0
• python35 branch
  • TensorFlow r0.12
  • numpy 1.11.0b1
  • scipy 0.18.1
  • scikit-learn 0.18.1
  • matplotlib 1.5.3
  • pandas 0.19.1
  • jupyter 1.0.0

Files Explained

• /capstone directory contains the code, notebooks, data, and report for the Udacity Machine Learning Nanodegree capstone project.

• /cifar10 directory stores the code and files for further exploring the CIFAR10 dataset using deep learning techniques after the Udacity Machine Learning Nanodegree capstone project.

• /mnist directory is a project for building a digit recoginizer upon MNIST dataset (incorporating kaggle playgorund competition) using machine learning and deep learning techniques. Additionally, various machine learning technique and architectures are explored using MNIST dataset: fractional max pooling, 1x1 network in network architecture, very basic inception architecture, k-fold validation, ensembling multiple models(bagging and stacking).

• /leaf directory is a project for (incorporating kaggle playground competition) building a leaf classifier using machine learning techniques.

• /studybn directory contains a few scripts for studying and testing bach normalization and a notebook for demonstrating how to properly use TensorFlow tf.contrib.layers.batch_norm.

• cnn.py and test_cnn.py: the former is a quick wrapper building on top of TensorFlow to enable an easier and faster usage for building a CNN model (note that only the commonly recommended/default model building options are included and implemented in cnn.py); the latter performs unit testing on the interfaces provided by cnn.py.

• preprocess.py: contains a set of functions for preprocessing image input samples: including centering data, performing PCA whitening, ZCA whitening, image darkening, image brightening, and random image flipping (up-down-flip and left-right-flip).