issamlaradji/neuralnetworks

Multi-layer Perceptron

Neural Network Toolbox

This repository contains neural networks implemented in Theano. Theano is a great optimization library that can compile functions and their gradients. It can also harness the GPU processing power if Theano is configured correctly.

The repository neural_network_theano can be used in a similar way as scikit-learn. But since neural networks can have different layers and update rules, initializing them is just a bit more complicated.

examples.py shows how to initialize a classification and a regression model and train them on different datasets.

References

• https://github.com/Theano/Theano
• https://github.com/HIPS/autograd
• https://github.com/scikit-learn/scikit-learn

neuralnetworks's Issues

About the gradient descent

Hi, I think your code is very useful. But 'l-bfgs' seems to out perform 'sgd' consistently, which seems counter-intuitive to me. One thing I have in my mind is for 'sgd' it does not include the momentum to accumulate the past gradients. I would like to add that into your code and maybe try to merge it to your code. Is that ok to you?

Add configuration instructions to README.md

Sorry - new to Python. How would I use this in a project? I can't seem to get the paths and import statements right.

About Multi-Layer Adoption of Code Design

Hello Mr. Laradji,

First of all, let me introduce myself. My name is Batuhan from Turkey. I am studying as M.Sc. student on Material Science & Engineering Department. My thesis topic is "Composition Optimization of Nb-included Al7SiMg alloy with ANN" even though I haven't experienced on ANN on Python. My experiment case is:

Inputs:
1- %Niobium amount (%0.005 for ex.)
2-%Titanium amount (%0.01 for ex.)
3-% Boron amount(%0.004 for ex.)

Outputs:
1-Yield Strength (190 MPa for ex.)
2-Tensile Strength (250 MPa for ex.)
3- % elongation (%5.4 for ex)

I will cast and obtain some of tensile test results from casting specimen with random Nb,Ti and B addition as test data. My aim is to train the algorithm with these test data and obtain a optimization function with >95 R^2 data.

How would I adopt your code into multi-output ANN model with same procedure? I cannot comprehend correctly how to put them into your open-source coding design. I would be pleased If you can help on this issue. As enthusiastic student, I would like to learn how to shed light on that kind of structure.
DIY ANN.zip

---