neural-monkey 0.2.0

A neural network library for beginners.

Monkey – Simple Neural Networks for Beginners

Monkey is a lightweight Python library for building and training simple neural networks.
It is designed for beginners to learn concepts like layers, activations, forward propagation, backpropagation, and simple attention mechanisms in a fun and interactive way.

---

Features

• Create fully connected neural networks (Dense layers)
• Choose activation functions: ReLU, Sigmoid, Tanh
• Train networks using gradient descent with adjustable learning rate and epochs
• Make predictions on new inputs
• Optional lightweight AttentionBlock for sequence inputs
• Beginner-friendly API with minimal setup
• Fully Python-based, no external dependencies

---

Installation

pip install monkey

---

Quick Start Examples

1. Predict the sum of two numbers

from monkey.nn import NeuralNet

# Training data
x_train = [[2, 8], [9, 3], [7, 4], [1, 1]]
y_train = [[sum(pair)] for pair in x_train]

# Create network
nn = NeuralNet(input_size=2)
nn.add_layer(neurons=5, activation='relu')
nn.add_layer(neurons=1, activation='relu', layer='output')

# Train the network
nn.train(x_train, y_train, epochs=500, lr=0.1)

# Predict
print("Prediction for [3,5]:", nn.predict([3,5])[0])

2. Predict using Sigmoid activation

nn = NeuralNet(input_size=2)
nn.add_layer(neurons=4, activation='sigmoid')
nn.add_layer(neurons=1, activation='sigmoid', layer='output')

nn.train(x_train, y_train, epochs=1000, lr=0.05)
print("Prediction for [2,2]:", nn.predict([2,2])[0])

3. Using AttentionBlock for sequences

from monkey.attention import AttentionBlock

seq_input = [[0.8, 0.2, 0.1], [0.5, 0.1, 0.3], [0.2, 0.7, 0.6]]
attn = AttentionBlock(input_size=3, output_size=3)
seq_output = attn.forward(seq_input)

print("Attention output:", seq_output)

---

API Reference

Class / Function	Description
NeuralNet	Core class for creating and training fully connected networks
NeuralNet.add_layer	Add a layer to the network; choose neurons, activation, layer type
NeuralNet.predict	Make predictions for a given input
NeuralNet.train	Train the network on input/output data
AttentionBlock	Simple attention mechanism for sequences
relu(x)	Rectified Linear Unit activation function
sigmoid(x)	Sigmoid activation function
tanh(x)	Tanh activation function

Layer Parameters in add_layer:

• neurons: Number of neurons in the layer
• activation: One of 'relu', 'sigmoid', 'tanh'
• layer: 'hidden' (default) or 'output'
• input_size: Optional, only needed for first layer

---

Testing Monkey

You can run the provided test script:

python tests/test.py

Sample test scenarios include:

• Predicting sums of two numbers
• Predicting multiplication or XOR patterns
• Testing different activation functions
• Testing AttentionBlock on sequence data

---

Learning Tips for Beginners

• Start with a single hidden layer and few neurons
• Use small datasets (like sum of two numbers) for testing
• Adjust learning_rate and epochs to see effect on convergence
• Observe how activation functions change the output and training speed
• Experiment with the AttentionBlock for sequence-based learning

---

Contributing

Monkey is beginner-friendly and open to contributions!

• Experiment with new activation functions
• Add utilities or visualization helpers
• Improve training efficiency

Github Repo available at : https://github.com/19919rohit/Neural-Monkey

---

License

MIT License