This repository provides source code for guide reading course

it's constructed as following:

In this tutorial, we are going to practice DNN to solve notMNIST dataset. This dataset is designed to look like classic MNIST dataset, while looking a little more like real data: it's a harder task, and the data is a lot less 'clean' than MNIST.

There are three parts in this tutorial:

• Download dataset and pre-processing for training and testing
• Construct DNN with gradient descent
• Construct DNN with stochastic gradient descent

For more information about MNIST, please reference here
For more information about notMNIST, please reference here
All of the source code is reference from here

The tuorial is modified from https://github.com/Hvass-Labs/TensorFlow-Tutorials

The major tutorial file is tutorial.ipynb

You may need to install ipython Notebook

In this turtorial, we will show composition of CIFAR10 dataset and some preprocess for input image at first.

Then, demostrate how to build up a CNN network and train it.

Finally, we will review the training results and some parameters attributes after training

In this tutorial, we are going to build LSTM character model. The dataset we use in this practice is Text8.

There are two parts in this tutorial:

• Create word2vector
• Build LSTM char-level model to generate text

For more information about Text8, please reference here

There are four main module in this solution

Input cifar10 dataset --> build model --> build loss function and log --> training

This script includes two network: ResNet and GoogleLeNet You can use model ticket to build these networks by calling: mz = model_zoo.build_model(#model ticket) network = mz.build_model()

Common module for building a CNN network.

The script for evaluation trained model on whole cifar10 test set