HyperFlow-Package 1.0.0

HyperFlow: Next-Generation Computational Framework for Machine Learning & Deep Learning

HyperFlow: Next-Generation Computational Framework for Machine Learning & Deep Learning

📄 Read the Full Documentation

🚀 Introduction

HyperFlow is an advanced computational framework designed to enhance machine learning and deep learning research. At its core is the FlowUnit class, an optimized and modular implementation supporting:

• 🔢 Mathematical operations
• ⚙️ Activation functions
• 📉 Optimization techniques
• 🔄 Backpropagation

HyperFlow provides simplicity and power, making it ideal for understanding neural networks, automatic differentiation, and optimization techniques.

🎯 Inspiration

Inspired by Micrograd by Andrej Karpathy, HyperFlow extends its capabilities by offering:

• 🚀 Advanced mathematical operations
• 🎯 Broader activation function support
• ⚡ Optimized backpropagation
• 🔍 Enhanced modularity

It serves as a lightweight yet powerful alternative to complex deep learning frameworks like PyTorch.

💡 Why HyperFlow?

✅ Lightweight & Transparent – Focuses on raw Python implementations to help understand ML/DL concepts.
⚡ Efficient & Optimized – Uses map functions for better performance.
🔧 Flexible & Powerful – Supports neural networks, including backpropagation.
📉 Minimal NumPy Dependency – Encourages learning without excessive reliance on pre-built libraries.

🔑 Core Functionalities

🔢 Mathematical Operations

• create2darray, convert2darray, add, sub, mul, matmul, dot, pow

⚙️ Activation Functions

• sigmoid, tanh, ReLU, Leaky ReLU, softmax

📉 Loss Functions

• categorical_cross_entropy, binary_cross_entropy, mse_loss

🔄 Optimization & Backpropagation

• backpropagate, gradient_descent

🧠 Neural Network Implementation

The Neuron.py module simplifies the creation of:

• 🏗 Neurons
• 🔗 Layers
• 🏛 Complete Neural Networks

This module offers full control over weights, biases, and architecture for in-depth experimentation.

📌 Examples & Outputs

✅ Dot Product Calculation

from hyperflow import FlowUnit

a = FlowUnit([1, 2, 3])
b = FlowUnit([4, 5, 6])

result = a.__dot__(b)
print(f"Dot product result: {result.data}")

Output:

Dot product result: 32

🔄 Backpropagation Test

def test_backpropagation():
    x = FlowUnit(2.0)
    y = FlowUnit(-3.0)
    z = FlowUnit(1.5)

    a = x.sigmoid()
    b = y.tanh()
    c = z.relu()
    d = x.leaky_relu()

    loss = a + b + c + d
    loss.backpropagate()

    print(f"x.grad: {x.grad}")
    print(f"y.grad: {y.grad}")
    print(f"z.grad: {z.grad}")

test_backpropagation()

Output:

x.grad: 1.1049935854035067
y.grad: 0.00986603716543999
z.grad: 1.0

🔢 Activation Function Tests

def test_flow_unit_functions():
    data = np.array([1.0, 2.0, 3.0])
    flow_unit = FlowUnit(data)

    sigmoid_out = flow_unit.sigmoid()
    print("Sigmoid Output:", sigmoid_out.data)

    tanh_out = flow_unit.tanh()
    print("Tanh Output:", tanh_out.data)

    relu_out = flow_unit.relu()
    print("ReLU Output:", relu_out.data)

    leaky_relu_out = flow_unit.leaky_relu(alpha=0.01)
    print("Leaky ReLU Output:", leaky_relu_out.data)

    softmax_out = flow_unit.softmax()
    print("Softmax Output:", softmax_out.data)

test_flow_unit_functions()

Output:

Sigmoid Output: [0.731 0.880 0.952]
Tanh Output: [0.761 0.964 0.995]
ReLU Output: [1.0 2.0 3.0]
Leaky ReLU Output: [1.0 2.0 3.0]
Softmax Output: [0.090 0.244 0.665]

📉 Loss Function Tests

def test_loss_functions():
    logits = FlowUnit(np.array([2.0, 1.0, 0.1]))
    target = [1, 0, 0]

    cce_loss = LossFunctions.categorical_cross_entropy(logits, target)
    print("Categorical Cross-Entropy Loss:", cce_loss.data)

    inputs = FlowUnit(np.array([[1.0, 2.0], [3.0, 4.0]]))
    target = FlowUnit(np.array([1, 0]))
    parameters = (FlowUnit(np.array([0.5, -0.5])), FlowUnit(0.1))

    bce_loss = LossFunctions.binary_cross_entropy_loss(inputs, target, parameters)
    print("Binary Cross-Entropy Loss:", bce_loss.data)

test_loss_functions()

Output:

Categorical Cross-Entropy Loss: 0.417
Binary Cross-Entropy Loss: [0.715]

📚 Explore More Use Cases

Find more examples and use cases in the GitHub repository.