ak-pynn 0.1.8

A simplistic and efficient pure-python neural network library that allows to build multilayer neural network with ease.

MultiLayer Neural Network (ak_pynn)

A simplistic and efficient pure-python neural network library, which can be used to build , visualize and deploy deep learning ANN models. It is optimized for best performance.

• Optimized for performance
• Better Visualization
• Cross platform

Authors

• @ankit_kohli

License

MIT

Support

For support, email [e-mail]

Features

• Efficient implementations of activation functions and their gradients

  • Sigmoid
  • ReLU
  • Leaky ReLU
  • Softmax
  • Softplus
  • Tanh
  • Elu
  • Linear

• Efficient implementations of loss functions and their gradients

  • Mean squared error
  • Mean absolute error
  • Binary cross entropy
  • Categorical cross entropy

• Several methods for weights initialization

  • 'random uniform', 'random normal'
  • 'Glorot Uniform', 'Glorot Normal'
  • 'He Uniform','He Normal'

• Neural network optimization using

  • Gradient Descent (Batch/ SGD / Mini-Batch)
  • Momentum
  • Adagrad
  • RMSprop
  • Adam

• Regularizations

  • L1 Norm
  • L2 Norm
  • L1_L2 Norm
  • Dropouts

• Batch Normalization

• Early Stopping

• Validation Splits

• Predict Scores

Installation

Install the release (stable) version from PyPi

pip install ak-pynn

Usage/Examples

Import

from ak_pynn.mlp import MLP

Usage

model = MLP()
model.add_layer(4,input_layer=True)
model.add_layer(10,activation_function='relu',batch_norm=True)
model.add_layer(10,activation_function='relu',dropouts=True)
model.add_layer(10,activation_function='relu')
model.add_layer(3,activation_function='softmax',output_layer=True)
model.compile_model(optimizer='Adam',loss_function='mse',metrics=['mse','accuracy'])

Output

                                ( MODEL SUMMARY )                        
        
        ===================================================================
               Layer           Activation    Output Shape      Params    
        ===================================================================

               Input             linear       (None, 4)          0       
        -------------------------------------------------------------------

               Dense              relu        (None, 10)         50      
        -------------------------------------------------------------------

         BatchNormalization       None        (None, 10)         40      
        -------------------------------------------------------------------

               Dense              relu        (None, 10)        110      
        -------------------------------------------------------------------

              Dropout             None        (None, 10)         0       
        -------------------------------------------------------------------

               Dense              relu        (None, 10)        110      
        -------------------------------------------------------------------

               Output           softmax       (None, 3)          33      
        -------------------------------------------------------------------

        ===================================================================

        Total Params  - 343
        Trainable Params  - 323
        Non-Trainable Params  - 20
        ___________________________________________________________________
              

Visualizing model

model.visualize()

Training the model

model.fit(X_train, Y_train,epochs=200,batch_size=32,verbose=False,early_stopping=False,patience=3,validation_split=0.2)
model.predict_scores(X_test,Y_test,metrics=['accuracy','precision','macro_recall'])
plt.plot(model.history['Val_Losses'])
plt.plot(model.history['Losses'])

TESTS

@mnist_test

@iris_test

@mlp_demo

Citation

If you use this library and would like to cite it, you can use:

Ankit kohli, "ak-pynn: Neural Network libray", 2023. [Online]. Available: https://github.com/ankit869/ak-pynn. [Accessed: DD- Month- 20YY].

or:

@Misc{,
  author = {Ankit kohli},
  title  = {ak-pynn: Neural Network libray},
  month  = May,
  year   = {2023},
  note   = {Online; accessed <today>},
  url    = {https://github.com/ankit869/ak-pynn},
}