alrojo / tensorflow-tutorial Goto Github PK

Practical tutorials and labs for TensorFlow used by Nvidia, FFN, CNN, RNN, Kaggle, AE

Python 7.13% Jupyter Notebook 92.87%

tensorflow-tutorial's Introduction

TensorFlow Tutorial - used by Nvidia

Learn TensorFlow from scratch by examples and visualizations with interactive jupyter notebooks. Learn to compete in the Kaggle leaf detection challenge!

All exercises are designed to be run from a CPU on a laptop, but can be accelerated with GPU resources.

Lab 1-4 was used in the Deep Learning using TensorFlow in London by Nvidia and Persontyle

Credits

Labs 1, 2, 3 and 5 have been translated from Theano/Lasagne with minor modifications from the following repositories: Nvidia Summer Camp and 02456 deep learning. Original authors: skaae, casperkaae and larsmaaloee.

Thanks to professor Ole Winther for supervision and sponsoring the labs.

Setup and Installation

Guides for downloading and installing TensorFlow on Linux, OSX and Windows using Docker can be found here.

Material

The material consists of 5 labs.

Lab1 - FFN

Logistic regression, feed forward neural network (FFN) on the (in)famous MNIST!

Optional reading material from Michael Nielsen chapters 1-4 (Do 3-5 of the optional exercises).

Lab2 - CNN

Convolutional Neural Network (CNN) and Spatial Transformer on images.

Optional reading material from Michael Nielsen chapter 6 (stop when reaching section called Other approaches to deep neural nets).

Lab3 - RNN

Recurrent Neural Network (RNN) on Translation using Encoder-Decoder model and Encoder-Decoder with attention.

Optional reading material from Alex Graves chapters 3.1, 3.2 and 4,

Lab4 - Kaggle

Compete in the kaggle competition Leaf Classification using FFN, CNN and RNN.

Lab5 - AE

Unsupervised learning with autoencoder (AE) reconstructing the MNIST from only two latent variables.

Optional reading material from deeplearningbook.org chapter 14.

tensorflow-tutorial's People

Contributors

Stargazers

Watchers

Forkers

zakiindrasukma yiweifang annajiat upstatepedro giorgospa robinbing tdy akansal1 vyraun bertomartin revbooyah ratulghosh ad05bzag arvindfaizal pepsalehi phpmind paddymahoney simujenni wanjinchang zhonghp wuqixiaobai lixiang0 gitkeidy lymrty mistobaan mingwen-cs c-rap backupmanager thesillystudent fsodogandji stephanos-stephani ccczone szabolcsmaj diwahars alitabet zkailinzhang qgzang dyhpoon qinghsui benjamesbabala jackoll8868 akivaron dhaneshr rhythm92 akshittyagi hari195 weilamchung xieguobin isrmehta paulsturgesscv manila95 kunlqt gersongams sayanddude tchangw drschilling lkngin vanova yaowee renyuan1010 anirbanniara andandandand shashankg7 tuevo21 bastinrobin jeperez aaronring kentchun33333 leeparayno milossimic jimmyyentran innovarul timesofbadri rayed-therap lyk125 youngdev ezhangle rogerborras projectafey clear-datacenter badlogicmanpreet mjk276 adonese prakhar-agarwal ompanda sjqgithub s4sarath bkjackson peterquid ibelieveai gth158a pengsongyou wubr2000 dima1034 ngoduyvu leopardpan sibosutd lunasdejavu vuongnm merryhunter

tensorflow-tutorial's Issues

tf.pack should be renamed to tf.stack

In lab2_CNN/lab2_CNN.ipynb, there is an AttributeError: 'module' object has no attribute 'pack'.
It should be renamed to tf.stack according to https://www.tensorflow.org/install/migration.

Lab4 Kaggle - argument typo

In lab4 "Build the model", convolutional2d is called as

# wrapping conv with batch_norm
def conv(l_in, num_outputs, kernel_size, scope, stride=1):
    return convolution2d(l_in, num_outputs=num_outputs, kernel_size=kernel_size,
                         stride=stride, normalize_fn=batch_norm, scope=scope)

Which produces:

TypeError: convolution2d() got an unexpected keyword argument 'normalize_fn'

It should be "normalizer_fn". Adding that missing R makes it works perfectly.

Docker

Hi alrojo,

I am sure that most people have had some experience with docker. I have never used it so I had some difficulty getting Tensorboard up with the following commands. I also checked out the man page, but it seemed that there were multiple options for one flag. Would you be able to expound on the following commands?

> docker run -p 6006:6006 -v $PATH\_TO\_FOLDER/tensorflow_tutorial:/mnt/myproject -it alrojo/tf-sklearn-cpu
> cd mnt/myproject/lab6_Kaggle
> tensorboard --logdir=tensorboard

Thanks again for the great tutorial!

Only folders available in container(shared directory)

I have cloned the repository to C:\Users\Mathias\Desktop\Docker so that i have C:\Users\Mathias\Desktop\Docker\tensorflow_tutorial\download-and-setup and so on.
if I run the following command:
docker run -p 8888:8888 -v C:\Users\Mathias\Desktop\Docker\tensorflow_tutorial:/mnt/myproject -it alrojo/tf-sklearn-cpu
I have a folder named tensorflow_tutorial within /mnt/myproject/ but no files.
If i change the folder for sharing i can see subdirectories but not still not the content of these.
I feel like the point of a shared directory is to be able to share files?

Print tails

I had a suggestion, but realized it was not a big deal. And I don't know how to delete my own issue.

Thanks a lot for the tutorial. I have learned so much!

Just a little typo


print("operations")
operations = [op.name for op in tf.get_default_graph().get_operations()]
print(operations)
print

print("variables")
variables = [var.name for var in tf.all_variables()]
print(operations)

The print on last line should be:
print(variables)

How to make prediction for new cases in lab3_RNN ?

Hi ,

Its more or like a pull request . You have defined both training and validation in the decoder itself . So , I know the idea behind making new prediction .

The problem with the validation inside tf_utils.decoder is , it is always expecting the target .
But for making a prediction ( without having targets , basically predicting the target ) , we need to have a separate function . How to write that ? I know , the idea , taking the first target as EOS , and then keep on finding the argmax until we reach max_length or another EOS . Can you help to make a function , doing the same ?