MLPApproximator

MLP Neural Network Learning

Educational project of function approximation. Main goal how increasing
number of hidden layer perceptron influence on accuracy and quality of approximation.

3 function are mapped into three inputs giving 3 outputs.

Getting Started

Installing

• python3.7
• numpy
• matplotlib - optional

python -m pip install -r requirements.txt
python MLPApproximatorConsoleUI.py -h

Example of usage

$> python MLPApproximatorConsoleUI.py -ds Examples/DataSetM5.txt -e 10
Approximator:   MLP Function Approximator
Approximator:   input number=3
Approximator:  output number=3
Approximator:  hidden number=3
Approximator:  Train on 82 samples
Approximator:  Epoch:    1/10
Approximator:   Epoch Time=0.0684s GlobalTime=0.0684s Loss=15.6%
Approximator:  Epoch:    2/10
Approximator:   Epoch Time=0.0408s GlobalTime=0.109s Loss=15.6%
(...)
Approximator:  Epoch:    9/10
Approximator:   Epoch Time=0.0731s GlobalTime=0.498s Loss=15.6%
Approximator:  Epoch:   10/10
Approximator:   Epoch Time=0.057s GlobalTime=0.555s Loss=15.6%
Approximator:   Training Time=0.555s
Approximator:  Testing:
Approximator:   Loss=15.8%

Running the tests

Whole research is included into integration test. The result are saves into ‘TestResults’ folder.
It takes around 1h per group. 3 groups exist.

python -m unittest MLPApproximatorTest.test_integration.TestIntegration

Author

• Kajetan Brzuszczak - Submitted - HalfInner

License

This project is licensed under the MIT License - see the LICENSE.md file for details