awjuliani / tf-tutorials

A collection of deep learning tutorials using Tensorflow and Python

#Tensorflow Tutorials This repository contains a collection of miscellaneous Jupyter notebooks which implement or provide a tutorial on a different Deep Learning topic. All models are implemented in Tesnorflow.

• DCGAN - An implementation of Deep Convolutional Generative Adversarial Network.
• InfoGAN An implementation of InfoGAN: Interpretable representation learning by information maximizing generative adversarial nets
• Deep Layer Visualization - Tutorial on visualizing intermediate layer activation during MNIST classification.
• Deep Network Comparison - Implementations of ResNet, HighwayNet, and DenseNet, for CIFAR10 classification.
• RNN-TF - Tutorial on implementing basic RNN in tensorflow.
• t-SNE Tutorial - Tutorial on using t-SNE to visualize intermediate layer representation during MNIST classification task.

tf-tutorials's Issues

In deep network comparison

Hello!
I followed the "Deep Network Comparison" tutorial: I put the DenseNet code block and all other code blocks (except for RegularNet, ResNet, and HighwayNet) into a python file, and tried to run it. However I see:

python densenet.py 
<IPython.core.display.HTML object>
Switched CIFAR set to 2
Step: 0 Loss: 2.64302 Accuracy: 0.109375
terminate called after throwing an instance of 'std::bad_alloc' 
  what():  std::bad_alloc

Did I run it in a wrong way? How should I solve this problem?

input_data import

Hi, your "import input_data" is not working for me. In the official TF tutorial, the import call changed to "from tensorflow.examples.tutorials.mnist import input_data".
Thx for the tuts, they are great!

Deep Network Comparison.ipynb cannot achive the accurity

In the readme of " Deep Network Comparison.ipynb"
With little parameter tuning I was able to get them to perform above 90% accuracy on a test set after only an hour or so.

The train phase had been pass 12 hours ,however the test set accuracy is so low.

Step: 400 Loss: 1.75564 Accuracy: 0.234375
Test set accuracy: 0.252

My environment:
hardware: i7 cpu, 980ti nvidia card , 24G
software: cuda7.5 ,tensorflow 1.1, python 3, cudnn5.0

ValueError

Got ValueError on my MacBook Pro with Python 2.7
ValueError: Filter must not be larger than the input: Filter: (3, 3) Input: (2, 2)

incompatibility and nan to report

First, in the infoGAN tutorial, the tf.concat and tf.split still uses the old ordering of arguments.
second, by running the code, after several iterations, the program generates NaN in Gen and Disc losses