nilesh-patil / digit-classification-fully-connected-neural-nets

Pure Numpy implementation of Fully Connected Neural Networks with gradient descent for weight optimisation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Structure :

The repository contains code for fully connected neural network & also a demo.py file used to define & run the network. A decent variety of tests are implemented and can be seen in the solution.ipynb notebook(end). The network is designed to classify MNIST digit recognition dataset.

  • ./code folder:

    • contains original files with implemented classes
    • Added tests_all.py file which contains code used for creating plots & testing variations in the parameters
  • report.pdf:

    • Contains report for the network performance on different parameters
  • Solution.ipynb contains the full standalone work :

    • Fully implemented classes
    • Demo run
    • Testing code
    • variation tests
    • Report generation python code & markdown
  • Solution.html is just the html version for this.

  • I have added my results.pkl file, which is essentially the object where I have stored results from testing.

About

Pure Numpy implementation of Fully Connected Neural Networks with gradient descent for weight optimisation


Languages

Language:HTML 63.3%Language:Jupyter Notebook 33.7%Language:Python 3.0%