Deep Learning Thesis Project

by: Apostolis Gioulis

The projects purpose is to create a work-flow using some of the available tools used in modern Deep Learning algorithm Development. To be specific, Theano and Torch7 libraries are used as well as a custom python implementation using Python.

Implementations

There are 3 folders containing 3 different implementations as well as some other modules of the supportive nature:

numpy/: Contains a NN implementation using python extended with the scientific computing library numpy. The implementation essentially is an extension of the code published by Michael Nielsen for education purposes. All is downloaded in the data/ folder located at the same level as the thesis root folder.
theano/: Contains a NN implementation written in Python using the theano framework. The implementation is also based on the work of M.Nielsen
Torch7: Contains a NN implementation using Torch7 framework as well as the dp library by Nicholas Leonard The code used is actually an example provided by Nicholas Leonard

Dependencies

numpy/: In order for the numpy implementation to run successfully the following python packages must be installed.
- numpy
- configparser
- clint
- requests
They can be installed by using the sudo pip install <package_name> command
theano/: In order for the theano implementation to run successfully: theano library must be installed. To do so either:
- Use the automated tool theano/theano_setup.sh developed for the purposes of the work-flow (tested in fedora21, ubuntu 14.04)
- Follow the instructions available in the official theano website
All of the aforementioned python packages are required as well.
Torch7: In order for the torch7 NN implementation to run torch7 must be installed in the users system. To do so:
- Use the automated tool Torch7/torch7_setup.sh that in turn uses the official installation tool. If the script does not work some debugging options are listed as comments in the script. The script in turn requires
  - cmake
  - curl
Once torch7 is installed the dp library must be installed as well (Torch7/torch7_setup.sh installs it automatically by executing)
```
 luarocks install dp --local
```

Execution

numpy/:
- Configure: By tampering with the numoy/src/network.ini file
- Run: ./run_network.py
theano/:
- Configure: By tampering with the theano/src/network.ini file
- Run: ./T_run_network.py
Torch7::
- Configure: There are many Configuration options by passing command line args. For more info on the available options read line 16-32 of the file
- Example Run: th simple_NN.lua --learningRate 0.25 --dataset Mnist

Licence

The MIT License (MIT)

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.