Note is a versatile machine learning library designed to streamline the construction and training of neural networks. It serves as a robust tool for both deep learning and reinforcement learning applications, allowing users to effortlessly train agents using Note, Keras, or PyTorch.

Key Features

Deep Learning

The Note.nn.layer package includes a variety of modular layer components that facilitate the building of neural networks. Built with TensorFlow compatibility in mind, these layer modules provide flexibility in network construction, accommodating both Keras-style and PyTorch-style architectures. With Note, defining the shape of training parameters can be done dynamically during data input, similar to Keras, or set in advance as in PyTorch.

Example Usage with PyTorch

from Note import nn

class model(nn.Model):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.dense(128, 784, activation='relu')
        self.layer2 = nn.dense(10, 128)
    
    def __call__(self, data):
        x = self.layer1(data)
        x = self.layer2(x)
        return x

Example Usage with Keras

from Note import nn

class model(nn.Model):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.dense(128, activation='relu')
        self.layer2 = nn.dense(10)
    
    def __call__(self, data):
        x = self.layer1(data)
        x = self.layer2(x)
        return x

Reinforcement Learning

Note simplifies the reinforcement learning process by allowing agents to be constructed by inheriting from the RL or RL_pytorch classes. Training, saving, and restoring agents created with Note or other frameworks is straightforward, enabling efficient learning.

Comprehensive Documentation

An extensive collection of documentation is available, covering usage, examples, and advanced functionalities. Key components include:

• Layer Documentation
• Model Documentation
• Reinforcement Learning Documentation
• Function Module Documentation

Additional Core Components

• Model Management: The Note.nn.Model.Model class manages neural network parameters and layer configurations.
• Parameter Initialization: Functions like Note.nn.initializer.initializer and Note.nn.initializer.initializer_ are responsible for setting up the parameters of the neural network, ensuring optimized training behavior.
• Sequential Models: The Note.nn.Sequential.Sequential class is used similarly to Keras and PyTorch's sequential models, promoting an intuitive approach to layer stacking.

Community and Support

Note is actively supported and contributions are welcomed through Patreon. Feedback and suggestions can be communicated via email at notedance@outlook.com.

Explore the full potential of machine learning with Note, where robust design meets user-friendly functionality.