Update: the EMNIST dataset is no longer hosted on my University server (for som.py). It has to be re-downloaded, sorted and saved. The train_inputs_path and test_inputs_path variables will need to be linked to the final .csv file.
Use Kohonen's algorithm to make a Self Organising Maps for 3 different datasets:
- RGB dataset
- Iris dataset
- EMNIST dataset
The principal goal of this project is the implementation of a Kohonen Network as an application, in order to demonstrate its usefulness and explain the concepts of Machine Learning, by means of a converging Self-Organising Map.
The aim is to provide an in-depth study of Kohonen’s algorithm, and present insights of its properties, by designing and implementing a complete and functional model.
To execute the attached scripts, Flask, NumPy, Pandas, Matplotlib are required as a framework.
$ pip3 install Flask
$ pip3 install NumPy
$ pip3 install Pandas
$ pip3 install Matplotlib
The following used libraries are natively pre-installed in Python, but are nonetheless listed below:
• argsparse
• sys
• datetime
If necessary, virtual environments can be used to keep the libraries installed for the entire working machine seperate from those simply required for a specific task. This ensures that the libraries for this project don’t get change or mix up with the development PC’s native Python installation.
$ pip3 install virtualenv
$ cd ~/myPath/EMNIST-Kohonen-SOM/
$ virtualenv myFolder
$ source myFolder/bin/activate
$ pip3 install myPackages
$ deactivate
The parameter of each model can be updated. The arguments can be individually viewed:
$ Python3 iris.py --help
Which will return:
Make a 2D map of a multidimensional input
optional arguments :
−h, −−help show this help message and exit −d, −−debug Print debug messages to stderr
−r RATE, −−rate RATE Choose learning rate (range: 0−1)
Samples commands:
$ Python3 RGB.py -d -r=0.8 -i=1000
$ Python3 iris.py -d -r=0.3
$ Python3 som.py -d-r=0.3 -iTr=100 -iTe=100 -t=d
Where:
-d is the debug flag
-i is the number of inputs
-r is the learning rate
-iTr is the number of training inputs
-iTe is the number of testing inputs
-t is the flag to choose between digits (d), letters (l), or combined (c)
Finally the project can be run by executing app.py on the terminal:
$ python3 app.py
*Running on http://127.0.0.1:5000/ (Press CTRL+C to quit)
And then on a browser simply navigate to: http://127.0.0.1:5000. The website is now viewable.
Before:
After:
On the entier EMNIST dataset (of 47 classes):
