Home

Documentation

This project involves building an Artificial Neural Network featuring Stochastic Gradient Descent and various optimisers. It has been developed in Julia Version 0.5.1-pre+55 (2017-02-13 09:11 UTC) Commit 8d4ef37, Linux 4.4.0.

Requirements

The program was tested in presence of the following:

• Julia 0.5

Installation

The program does not require any installation. It can be run after cloning

git clone https://github.com/americast/ANN.git

Using the program

cd ANN  
julia ann.jl

Algorithms

• Stochastic Gradient Descent

After starting the program, one needs to enter the name of the input and output datasets.

input.txt
output.txt

Next, one needs to enter the number of nodes in the first hidden layer, followed by the second hidden layer, and so on. Here, 0 acts as a delimiter. For instance, if the first layer is to have 10 nodes, second layer 9 and third layer 10, one needs to enter

10  
9  
10  
0  

Next, one needs to enter the number of iterations to be performed.

200

Finally, one needs to enter a constant learning rate and momentum rate to be used.

1.15
0.5

After computation, the user will be prompted to test out the algorithm manually. For this the number of tests to be performed is to be entered followed by the input value of each test.

1
0 0 0

A prototype has been written in the file try which can be directly used as

julia ann.jl < try

The output in the prototype is one-hot encoded.

• Momentum optimiser

This is to be adjusted using the momentum rate as mentioned above.