nntool

Artificial Neural Network(ANN) Freamwork in python3 to read ANN topology from txt file

Disclaimer:

This repo is part of the undergraduate thesis of Mr. Efthymis Michalis. The thesis was developed under the supervision of Assistant Prof. Aggelos Pikrakis, in the Department of Informatics, School of ICT, University of Piraeus, Greece.

• Package Requirements

  1. docopt
  2. tensorflow
  3. scipy
  4. matplotlib
  5. pandas
  6. scikit-image
  7. numpy

Topology Syndax

Base Layer syndax: LayerType(param0,param1,...); Layer Types:

1. INPUT : Input(X_size,Y_size,Z_size)
2. CONVOLUTIONAL : Conv(block_size,Z_size,Act. Function)
3. POOLING : Pool(block_size,PoolMethod)
4. FULL CONNECTED : Fc(X_size,Act. Function)
5. DROPOUT : Dropout(probability%)
6. BATCH NORMALIZATION : BatchNorm()

Syndax:

1. INPUT must be the first layer

• X_size,Y_size,Z_size >= 1

2. CONVOLUTIONAL Act. Function:

• sigmoid
• tanh
• relu
• linear

4. POOL PoolMethods:

• max
• min
• avg

5. FULL CONNECTED Act. Function:

• All CONVOLUTIONAL Act. Functions
• softmax

DROPOUT

• probability between [0,100]

IRiS Dataset Example:

Input(4,1,1);
Fc(1024,tanh);
Fc(2056,tanh);
Fc(512,tanh);
Fc(3,softmax);