Skip to content

Latest commit

 

History

History
214 lines (162 loc) · 7.24 KB

README.md

File metadata and controls

214 lines (162 loc) · 7.24 KB
Raw

neural_network_in_python

Neural Networks: in C Script

Documentation for Neural Network Project

Overview

This documentation provides a comprehensive guide for using two Python scripts: neural_network.py for training a neural network model, and run_inference.py for running inference with a trained model.

Download Commands

git clone https://github.com/mdriyadkhan585/neural_network_python

Installing Dependencies

  1. requirements.txt

This file lists all the Python packages your project depends on. You can create it with the following content:

numpy

After cloning the repository, navigate into the project directory and install the dependencies listed in requirements.txt:

cd neural_network_python
pip install -r requirements.txt

neural_network.py - Model Trainer

Description

neural_network.py is responsible for defining, training, and saving a neural network model. This script includes:

  • Model architecture and activation functions
  • Weight initialization
  • Forward pass and backpropagation algorithms
  • Training loop with loss calculation and early stopping
  • Model saving

Key Functions

  1. initialize_weights(size)

    • Purpose: Initializes weights for the neural network layers.
    • Input: size (int) - Number of weights to initialize.
    • Output: np.ndarray - Array of initialized weights.

  2. relu(x)

    • Purpose: Applies the ReLU activation function.
    • Input: x (float) - Input value.
    • Output: float - Activated value.

  3. elu(x)

    • Purpose: Applies the ELU activation function.
    • Input: x (float) - Input value.
    • Output: float - Activated value.

  4. sigmoid(x)

    • Purpose: Applies the sigmoid activation function.
    • Input: x (float) - Input value.
    • Output: float - Activated value.

  5. swish(x)

    • Purpose: Applies the Swish activation function.
    • Input: x (float) - Input value.
    • Output: float - Activated value.

  6. softmax(output)

    • Purpose: Applies the softmax function to the output layer.
    • Input: output (np.ndarray) - Array of logits.
    • Output: np.ndarray - Softmax probabilities.
forward_pass(input, weights_hidden1, weights_hidden2, weights_hidden3, weights_output, biases_hidden1, biases_hidden2, biases_hidden3, biases_output)
  • Purpose: Performs the forward pass of the network.
  • Inputs:
    • input (np.ndarray) - Input data.
    • weights_hidden1, weights_hidden2, weights_hidden3, weights_output (np.ndarray) - Weights for each layer.
    • biases_hidden1, biases_hidden2, biases_hidden3, biases_output (np.ndarray) - Biases for each layer.
  • Outputs:
    • hidden1_output, hidden2_output, hidden3_output, final_output (np.ndarray) - Output of each layer.
backpropagate(input, hidden1_output, hidden2_output, hidden3_output, final_output, actual_output, weights_hidden1, weights_hidden2, weights_hidden3, weights_output, biases_hidden1, biases_hidden2, biases_hidden3, biases_output, learning_rate)
  • Purpose: Updates weights and biases through backpropagation.
  • Inputs:
    • input, hidden1_output, hidden2_output, hidden3_output, final_output (np.ndarray) - Layer outputs.
    • actual_output (np.ndarray) - True output values.
    • weights_hidden1, weights_hidden2, weights_hidden3, weights_output (np.ndarray) - Weights to be updated.
    • biases_hidden1, biases_hidden2, biases_hidden3, biases_output (np.ndarray) - Biases to be updated.
    • learning_rate (float) - Learning rate for updates.
  • Output: None.
    train(inputs, targets, num_samples)
  • Purpose: Trains the neural network model.
  • Inputs:
    • inputs (np.ndarray) - Input data.
    • targets (np.ndarray) - Target output data.
    • num_samples (int) - Number of training samples.
  • Output: None (Trains and saves the model).

How to Use

  1. Prepare Data: Organize your input data and target labels.
  2. Configure Parameters: Adjust the hyperparameters and model architecture if needed.
  3. Run Training: 
    python neural_network.py
    This will train the model on the provided data and save it to trained_model.pkl.

Example

Here's an example of the data preparation and training:

import numpy as np

# Sample data
inputs = np.random.rand(10, INPUT_SIZE)  # Replace with actual data
targets = np.eye(OUTPUT_SIZE)[np.random.randint(0, OUTPUT_SIZE, 10)]  # Example one-hot encoded targets

# Train the model
train(inputs, targets, len(inputs))

run_inference.py - Run Inference

Description

run_inference.py is used to load a pre-trained model and perform inference on new data. This script includes:

  • Model loading from a file
  • Forward pass for inference

Key Functions

load_model(weights_hidden1, weights_hidden2, weights_hidden3, weights_output, biases_hidden1, biases_hidden2, biases_hidden3, biases_output)
  • Purpose: Loads model weights and biases from a file.
  • Inputs:
    • weights_hidden1, weights_hidden2, weights_hidden3, weights_output (np.ndarray) - Arrays to store weights.
    • biases_hidden1, biases_hidden2, biases_hidden3, biases_output (np.ndarray) - Arrays to store biases.
  • Output: None.
forward_pass(input, weights_hidden1, weights_hidden2, weights_hidden3, weights_output, biases_hidden1, biases_hidden2, biases_hidden3, biases_output, hidden1_output, hidden2_output, hidden3_output, final_output)
  • Purpose: Performs inference by running a forward pass with the provided input.
  • Inputs:
    • input (np.ndarray) - Input data for inference.
    • weights_hidden1, weights_hidden2, weights_hidden3, weights_output (np.ndarray) - Loaded weights.
    • biases_hidden1, biases_hidden2, biases_hidden3, biases_output (np.ndarray) - Loaded biases.
  • Outputs:
    • hidden1_output, hidden2_output, hidden3_output, final_output (np.ndarray) - Output of each layer.

How to Use

  1. Load the Model: Ensure that trained_model.pkl is available in the same directory.

  2. Run Inference:

    python run_inference.py

Example

Here’s how to use the run_inference.py script to perform inference:

import numpy as np

# Load the trained model and perform inference
test_input = np.random.rand(INPUT_SIZE)  # Replace with actual test data
forward_pass(test_input, weights_hidden1, weights_hidden2, weights_hidden3, weights_output,
             biases_hidden1, biases_hidden2, biases_hidden3, biases_output,
             hidden1_output, hidden2_output, hidden3_output, final_output)

print("Inference result:")
for i in range(OUTPUT_SIZE):
    print(f"Class {i}: {final_output[i]:.4f}")

Summary

  • neural_network.py: For training the model and saving it.
  • run_inference.py: For loading the trained model and running inference.

Ensure that all dependencies (e.g., NumPy) are installed and the model file (trained_model.pkl) is available when running inference.