Hi, NumpyDL

NumpyDL is a simple deep learning library based on pure Python/Numpy. NumpyDL is a work in progress, input is welcome. The project is on GitHub.

The main features of NumpyDL are as follows:

• Pure in Numpy and native to Python
• Support basic automatic differentiation
• Support commonly used models, such as MLP, RNNs, GRUs, LSTMs and CNNs
• Perfect documents and easy to learn deep learning knowledge
• Flexible network configurations and learning algorithms.
• API like Keras deep learning library

The design of NumpyDL is governed by several principles:

• Simplicity: Be easy to use, easy to understand and easy to extend, to facilitate use in research. Interfaces should be kept small, with as few classes and methods as possible. Every added abstraction and feature should be carefully scrutinized, to determine whether the added complexity is justified.
• Transparency: Native to Numpy, directly process and return Python/Numpy data types. Do not rely on the functionality of Theano, Tensorflow or any such deep learning frameworks.
• Modularity: Allow all parts (layers, regularizers, optimizers, …) to be used independently of NumpyDL. Make it easy to use components in isolation or in conjunction with other frameworks.
• Focus: “Do one thing and do it well”. Do not try to provide a library for everything to do with deep learning.

User Guides

The NumpyDL user guide explains how to install NumpyDL, how to build and train neural networks using NumpyDL, and how to contribute to the library as a developer.

• 1. Installation
  • 1.1. Prerequisites
  • 1.2. Stable NumpyDL release
  • 1.3. Development installation
• 2. Development
  • 2.1. Philosophy
  • 2.2. What to contribute
  • 2.3. How to contribute

Tutorials

This is the tutorials of Deep Learning.

• 1. Activation
  • 1.1. What is activation function?
  • 1.2. Why need activation function?
  • 1.3. Commonly used activation functions
  • 1.4. What activation should I use?
• 2. Initialization
  • 2.1. Introduction
  • 2.2. Xavier Initialization
• 3. Objective
  • 3.1. What is the Objective Function?
  • 3.2. Visualizing the Objective function
• 4. Optimizer
  • 4.1. Gradient descent variants
  • 4.2. Challenges
  • 4.3. Parallelizing and distributing SGD
  • 4.4. Additional strategies for optimizing SGD
  • 4.5. Conclusion
• 5. Multilayer Perceptron
  • 5.1. Sigmoid function
  • 5.2. Back Propagation
  • 5.3. Example
  • 5.4. Code
• 6. Convolution Neural Networks
  • 6.1. Introduction
  • 6.2. Back Propagation
  • 6.3. Visualizing Features
  • 6.4. Code
• 7. Recurrent Neural Networks
  • 7.1. What is a Recurrent Neural Net?
  • 7.2. Simulating a Recurrent Neural Network
  • 7.3. Training a RNN - Backpropagation Through Time
  • 7.4. The Problems with Deep Backpropagation
  • 7.5. Long Short Term Memory
  • 7.6. Conclusions

API References

If you are looking for information on a specific function, class or method, this part of the documentation is for you.

• npdl.layers
  • Base Layers
  • Core Layers
  • Convolution Layers
  • Embedding Layer
  • Normalization Layer
  • Pooling Layers
  • Recurrent Layers
  • Shape Layers
• npdl.activations
  • Activations
  • Detailed Description
• npdl.initializations
  • Initializers
  • Detailed Description
• npdl.objectives
  • Examples
  • Objectives
  • Detailed Description
• npdl.optimizers
  • Examples
  • Optimizers
  • Detailed Description
• npdl.model
  • Detailed Description
• npdl.utils
  • Data Utils
  • Random Utils

Indices and tables

• Index
• Module Index
• Search Page