NeuralX

NeuralX is a Python package for creating neural networks with a wide range of features, including gradient checks, regularization, dropout, mini batch, and various optimization methods. This package is designed to be beginner-friendly, making it easy for users to understand neural network structures and build powerful models.

Features

• Gradient Checks: Ensure the correctness of your neural network's gradients with built-in gradient check tools.
• Regularization: Apply L1 and L2 regularization to prevent overfitting and improve model robustness.
• Dropout: Implement dropout layers for more effective regularization and enhanced generalization.
• Optimization Methods: Choose from different optimization methods, including the popular Adam optimizer, to train your models efficiently.

Installation

You can install NeuralX using pip:

pip install neuralx

Usage

To get started with NeuralX, follow these steps:

1. Import the necessary modules from NeuralX.
2. Build your neural network architecture, specifying layers, activation functions, and other parameters.
3. Train your model on your dataset using the chosen optimization method.
4. Evaluate the model's performance using the provided tools, including the confusion matrix implementation.

Here's a simple example of how to use NeuralX to create a basic neural network:

import neuralx as nx

# create layers description
layers = [
        (training_set.shape[0], None, None, None),
        (5, 'he', 'relu', 1),  # no_of_units, normalization_method, activation_function, dropout_keeping_probability
        (2, 'he', 'relu', 1),
        (1, 'he', 'sigmoid', 1)
    ]

# Create a neural network
model = nx.NeuralNetwork(layers)

# Train the model
model.train(
    training_set,
    training_set_labels,
    no_of_epochs=1000,
    optimization={'name': 'momentum', 'beta': 0.9},
    is_mini_batch=True,
    learning_rate=0.001
)

# Evaluate the model using the confusion matrix
cm = nx.ConfusionMatrix(model, test_set, test_set_labels)
print(cm)
print(cm.statistics())

provided training_set, training_set_labels, test_set and test_set_labels the result should be like this

          Total = 300 | Predicted Positive = 145 | Predicted Negative = 155
---------------------------------------------------------------------------
Actual Positive = 150 |           123            |            27           
---------------------------------------------------------------------------
Actual Negative = 150 |            22            |           128           


True Positive Rate (Sensitivity)(Recall): 0.82
                     False Negative Rate: 0.18
          False Positive Rate (Fall-out): 0.147
        True Negative Rate (Specificity): 0.853
   Positive Predictive Value (Precision): 0.848
                     False Omission Rate: 0.174
                    False discovery Rate: 0.152
   Negative Predictive Value (Precision): 0.826
                                Accuracy: 0.837
                       Balanced Accuracy: 0.836
                                F1 Score: 0.834
        Matthews Correlation Coefficient: 0.673
                   Fowlkes-Mallows Index: 0.834
                           Jaccard Index: 0.715
               Positive Likelihood Ratio: 5.578
               Negative Likelihood Ratio: 0.211
                   Diagnostic Odds Ratio: 26.436
                              Prevalence: 0.5
                    Prevalence Threshold: 0.297
                            Informedness: 0.673
                              Markedness: 0.674

Contributing

We welcome contributions from the community. Feel free to submit bug reports, feature requests, or even pull requests. Together, we can make NeuralX even better!

License

This project is licensed under the MIT License. See the LICENSE file for details.