This course, as part of [Summer Seminar ETSETB TelecomBCN, 4-8 July 2016 (http://telecomBCN.DeepLearning.Barcelona)] (http://telecomBCN.DeepLearning.Barcelona) is basically a hands-on tutorial that provides a quick start to building applications using TensorFlow. and we will teach the esential ideas of Tensorflow ecosystem.
- Maurici Yagües, Research engineer at BSC-CNS
- Jordi Torres, Professor at UPC and Researcher at BSC-CNS
We assume that the student has some basic knowledge about Python. If not, a Python Quick Start hands-on that will help to start with this language can be found here (Python Quick Start).
We assume that the student has a Ubuntu/Linux 64-bit or Mac OS X environment. If the student have a Windows environment we suggest to use VirtualBox in order to run a Linux in a separate virtual machine. You can follow the hands-on RUN A LINUX OS IN A VIRTUAL MACHINE to install it from this web.
Course grade are comprised of 3 homework assignments by groups (30%), class exercises (20%) and individual class attendance (50%).
- How to build basic TensorFlow graphs and how to train models
- Case study: Linear Regression in TensorFlow
- Basic data structures in TensorFlow
- Case study: Clustering in TensorFlow
- Single Layer Neural Network in TensorFlow
- TensorBoard
- Convolutional Neural Networks in TensorFlow
- TensorFlow High Level APIs: SLIM
- Recurrent Neural Networks in TensorFlow
We will use the book [First Contact with TensorFlow] (http://www.jorditorres.org/first-contact-with-tensorflow-book/) as a basic documentation. You can acces a [freely available on-line copy] (http://www.jorditorres.org/first-contact-with-tensorflow/>). The slides used during the hands-on will be also available before start the course. Additional documentation will be distributed during the course.
The slides and codes used during the sessions will be posted/updated 2 hours before the session:
- SLIDES 3 (pdf)
- SingleLayerNeuralNetwork.py
- input_data.py
- regression_tb.py (warning: use Google Chrome as a browser)
- regression_tb_md.py
- SLIDES 4 (pdf)
- MultiLayerNeuralNetwork.py
- slim_contrib.py (requires TF version 0.9)
- SLIDES 5 (pdf)
- rnn.py (requires TF version 0.9)
For the sessions, please bring your laptop, and you should have a working installation of Python. TensorFlow has a Python API (plus a C / C ++) that requires the installation of Python 2.7. Nowadays many Linux and UNIX distributions include a recent Python.If this is not the case I assume that any student who take this course knows how to install it from the general download page.
During the sessions lab the instructor could use IPython/Jupyter. If you are interested to use too, you can obtain it from [here] (https://ipython.org) (optional).
We will use a virtual environment virtualenv, a tool to create isolated Python environments to install TensorFlow. This will not overwrite existing versions and dependencies (and indirectly permissions) of Python packages from other projects required by TensorFlow in your laptop. Virtualenv creates an environment that has its own installation directories, that doesn’t share libraries with other virtualenv environments (and optionally doesn’t access the globally installed libraries either).
Follow the indications from the TensorFlow web page for installing the latest version of TensorFlow in a virtualenv.
The exemples in this hands-on will require install the following packages too:
$ sudo pip install numpy
$ sudo pip install matplotlib
In order to be sure that everything is working fine, create a simple TensorFlow code and save it with extension ".py". I suggest to use the following code multiplication.py from the course github:
import tensorflow as tf
a = tf.placeholder("float")
b = tf.placeholder("float")
y = tf.mul(a, b)
sess = tf.Session()
print sess.run(y, feed_dict={a: 3, b: 3})
You can download it from the github using the git command:
(telecomBCN)$ git clone https://github.com/jorditorresBCN/FirstContactWithTensorFlow.git
To run the code, it will be enough with the command
(telecomBCN)$ cd FirstContactWithTensorFlow
(telecomBCN)$ python multiplication.py
If the result is 9.0, it means that TensorFlow is properly installed.
Finally, when you’ve finished, you should disable the virtual environment as follows:
(telecomBCN)$ deactivate