LightNeuNet

A light-weight neural network framework built for academic use.

Credits

LightNeuNet borrowed syntaxes from Keras.

An example for Artificial Neural Network (ANN)

from lnn.model import SequentialModel
from lnn.layers import Input, Dense, Output

# Initialize a fully connected neural network
model = SequentialModel()

# Add the input layer
model.add(Input(32))

# Add the first and second hidden layer
model.add(Dense(18, activation='tanh', 
    kernel_initializer='uniform', use_bias = True))
model.add(Dense(18, activation='tanh', 
    kernel_initializer = 'uniform', use_bias = True))

# Add the output layer
model.add(Output(num_output_units, activation = 'softmax', 
    use_bias = True))

# Compile the model
model.compile()

# The summary of the model built
model.summary()

# Train the model
for i in range(5000):
    model.fit(X_train, y_train, learning_rate = 0.01)
    
y_pred = model.predict(X_test)

An example for Neural Networks optimized with Genetic Algorithm

from lnn.genetic_algorithm.layers import GAInput, GADense, GAOutput
from lnn.genetic_algorithm.model import GAModel, GASequentialModel

# Create the model
model = GAModel(population=500)

# Add input layer to the model
model.add(GAInput(32))

# Add the first hidden layer to the model
model.add(GADense(24, activation='tanh', 
    use_bias=True, kernel_initializer='uniform'))

# Add the second hidden layer to the model
model.add(GADense(18, activation='tanh', 
    use_bias=True, kernel_initializer='uniform'))

# Add the output layer to the model
model.add(GAOutput(4, activation='softmax', use_bias=False))

# Generate the population
model.new_population()

def play_snake(model, params=(...)):
    # Code to play the snake
    ...

iters = 10000
for i in range(iters):
    model.simulate(play_snake, keep_rate=0.6, 
        mutate_rate=0.01, params=(False,))