README

=== prerequisites ===

The code was build under Linux with Python 2.7.6 installed

Besides that you will need:

>> numpy, scipy, matplotlib, scikit-learn, scikit-image, cPickle, pybrain <<
The dependencies are also given per script below.

You can install all of the above in a Debian/Ubuntu system with:

sudo apt-get install build-essential python-dev python-numpy python-scipy python-matplotlib python-setuptools python-sklearn python-skimage

In order to run Rprop you will need pyBrain framework. Installation instructions here http://pybrain.org/docs/quickstart/installation.html

Note that if the above fails (IT SHOULDN'T), one can install the following packages using the python pip installer
sudo apt-get install python-pip
pip install <package_name_same as above examples>


1) Preprocess 

First of all preprocess.py script must be run in order to ensure that dataset is processed and placed in the right place.
If initial dataset does not exist, preprocess.py downloads it from the internet.

Execution:
>>> python preprocess.py


2) MLP:

Execution:
>>> python run_MLP.py --help

REMIND that: You can stop the execution at any time pressing CTRL-C, the object is saved and info is printed

optional arguments:
  -h, --help            show this help message and exit
  -t TRAIN, --train TRAIN
                        train function to use Back-propagation or Resilient
                        BackPropagation (B/R) ,default=B
  -l LAYERS, --layers LAYERS
                        number of layers (1 or 2 only implemented)
  -n1 NODES1, --nodes1 NODES1
                        number of 1st layer nodes, (keep it low or go
                        vacation)
  -n2 NODES2, --nodes2 NODES2
                        number of 2nd layer nodes, (keep it low or go
                        vacation)
  -du DUSAGE, --dusage DUSAGE
                        percentage of dataset used (default=10), enter 100 for
                        full dataset

if no arguments are supplied defaults values are used (Backprop - 1 layer - 50 nodes)

BackProp makes use of mlp.py and mlp2.py 
Rprop needs pyBrain libary installed.  http://pybrain.org/docs/quickstart/installation.html



3) KNN:

there is not a main function , parameters are defined on top of run_knn.py

Execution:
>>> python run_knn.py



4) SVM:
there is not a main function , parameters are set inside the script

Execution:
>>> python run_svm.py



5) CLASSIFY SCRIPT

python  classify.py --help
usage: classify.py [-h] -n ANN -p PICTURE [-fc FOREGROUND]

Classify a single number

  -h, --help            show this help message and exit
  -n ANN, --ann ANN     give filename where the trained ANN has been stored.
                        Required
  -p PICTURE, --picture PICTURE
                        give file path containing picture. Required
  -fc FOREGROUND, --foreground FOREGROUND
                        give foreground color (B/W), default:B

-n, -p : MANDATORY arguments

example:

>>>> python classify.py -n selected_instances/NN_100_0.save -p samples/7e.jpg

Results not so good, not sure if the right instances are saved. The 300_100 not working (confusion didnt have time to retrain it)