deeplearningkit 0.0.2

Implementation of neural network

DeepLearningKit

Description:

Welcome to DeepLearningKit, a Python library crafted with a passion for learning and a drive to simplify the journey into the depths of neural networks. As my inaugural project in this domain, DeepLearningKit is designed primarily for my educational purposes, aiming to help me to understand and experiment with deep learning concepts. This release marks the first version of DeepLearningKit, but the journey doesn't end here! I'll be actively working on improving and expanding the library regularly. Stay tuned for updates and new features in future releases.

Getting Started:

1. Installation:

   pip install deeplearningkit
   

2. Usage:

    import deeplearningkit as nn
    import numpy as np
    from preprocess import get_data, create_data
   
    # Load and create synthetic data (not in the kit)
    (train_X, train_Y), (test_X, test_Y) = create_data(100, 2)
   
    # Define and compile the neural network model
    model = nn.Model()
    model.add(nn.Layer.Dense(train_X.shape[1], 24, nn.Initializer.HeUniform(train_X.shape[1])), nn.Activation.ReLU())
    model.add(nn.Layer.Dense(24, 24, nn.Initializer.HeUniform(24)), nn.Activation.ReLU())
    model.add(nn.Layer.Dense(24, 2, nn.Initializer.HeUniform(24)), nn.Activation.Softmax())
    model.compile(nn.Optimizer.Adam(), nn.Loss.CategoricalCrossEntropy())
   
    # Train the model
    model.fit(x=train_X, y=train_Y, batch_size=None, epochs=2001, shuffle=True, display=True, plot=False)
   
    # Evaluate the trained model
    model.evaluate(test_X, test_Y)
   
    # Alternatively, load data directly from a CSV file using a preprocessing function (not in the kit)
    (train_X, test_X), (train_Y, test_Y) = get_data("data.csv")
   
    # Define and compile the model using a simpler function call
    model = nn.Model()
    model.add(nn.Layer.layer("dense", train_X.shape[1], 24, nn.Initializer.initializer("Heuniform", train_X.shape[1])), nn.Activation.activation('relu'))
    model.add(nn.Layer.layer("dense", 24, 24, nn.Initializer.initializer("heuniform", 24)), nn.Activation.activation("ReLU"))
    model.add(nn.Layer.layer("dense", 24, 1, nn.Initializer.initializer("henormal", 24)), nn.Activation.activation("Sigmoid"))
    model.compile(nn.Optimizer.optimizer("adam"), nn.Loss.loss("BinaryCrossEntropy"))
   
    # Train the model
    model.fit(x=train_X, y=train_Y, batch_size=None, epochs=1001, shuffle=True, display=True, plot=False)
   
    # Evaluate the model
    model.evaluate(test_X, test_Y)
   
    # Load a model configuration from a JSON file
    model_data = nn.parse_model_json("model.json")
   
    # Compile and fit the model using parsed data and a specified data preprocessing function
    model: nn.Model = nn.compile_and_fit_parsed_model(model_data, get_data, display=True, plot=False)
    model.evaluate(test_X, test_Y)
   
    # Or load, compile, fit, and evaluate directly from a JSON configuration
    model_data = nn.parse_model_json("model.json")
    model: nn.Model = nn.compile_fit_evaluate_parsed_model(model_data, get_data, display=True, plot=False)
   
    # Example of JSON configuration for a neural network model
    # {
    #     "loss_function": "BinaryCrossEntropy",
    #     "layers": [
    #         {"type": "Dense", "units": 24, "activation": "ReLU", "weights_initializer": {"type": "heUniform", "params": {"n_inputs": 2}}},
    #         {"type": "Dense", "units": 24, "activation": "ReLU", "weights_initializer": {"type": "heUniform", "params": {"n_inputs": 24}}},
    #         {"type": "Dense", "units": 1, "activation": "Sigmoid", "weights_initializer": {"type": "xavier", "params": {"n_inputs": 24, "n_outputs": 1}}}
    #     ],
    #     "epochs": 1001,
    #     "batch_size": 0,
    #     "optimizer": {"type": "Adam", "params": {}},
    #     "data": "data.csv"
    # }
   

License: DeepLearningKit is licensed under the MIT License.

Acknowledgments: This project wouldn't have been possible without the support and guidance from the deep learning community, bigup to the book : 'Neural Network From Scratch' by Harrison Kinsley and Daniel Kukiela <3 . Let's continue to learn!