🌟 Implementation of Neural Networks from Scratch Using Python & Numpy 🌟

Uses Python 3.7.4

• Core Concepts
• Activation Functions
• Loss Functions
• Optimizers

git clone <url>
pip install -r requirements.txt

Here, Keras is used just to load the MNIST dataset

• Tune hyperparameters in config.py
• Run the following command

python main.py

$ python main.py
epoch 1/30      error=0.173172
epoch 2/30      error=0.077458
epoch 3/30      error=0.058955
epoch 4/30      error=0.048161
.....
.....
.....
epoch 26/30     error=0.010333
epoch 27/30     error=0.009944
epoch 28/30     error=0.009602
epoch 29/30     error=0.009298
epoch 30/30     error=0.009045

Predicted Values: 
[array([[-4.31825197e-04, -1.80361575e-03,  6.84263430e-03,
        -1.42045839e-02, -1.32599433e-02, -3.67077777e-02,
         3.73258781e-02,  0.97446495,  4.59079629e-02,
        -8.94465105e-03]]), 
array([[ 0.0461294 , -0.00845601,  0.8578162 , -0.00272202,  0.01397735,
         0.17131938,  0.21350745, -0.06529926,  0.01975232, -0.10840968]])]
True Values: 
[[0. 0. 0. 0. 0. 0. 0. 1. 0. 0.]
 [0. 0. 1. 0. 0. 0. 0. 0. 0. 0.]]

• Activation Functions
  • Linear
  • Sigmoid
  • Tanh
  • Tanh
  • ReLu
  • LeakyReLu
  • SoftMax
  • GeLu
• Loss Functions
  • MAE
  • MSE
  • CrossEntropy
• Optimizers Functions
  • Gradient Descent
  • Gradient Descent w/ Momentum
  • Nestrov's Accelerated
  • RMSProp
  • Adam
• Regularization
  • L1
  • L2
  • Dropout
• Layer Architecture
• Wrapper Classes
• Hyperparameters Configuration
• Clean Architecture
• UI (Similar to Tensorflow Playground)

• Deep Learning Specialization, Andrew NG - Coursera
• Machine Learning Cheatsheet

© 2020 Ryan Dsilva