Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Single Path One-Shot NAS MXNet implementation with full training and searching pipeline. Support both Block and Channel Selection. Searched models better than the original paper are provided.

Stars: ✭ 136 (-24.44%)

Mutual labels: neural-architecture-search

Petridishnn

Code for the neural architecture search methods contained in the paper Efficient Forward Neural Architecture Search

Stars: ✭ 112 (-37.78%)

Mutual labels: neural-architecture-search

Dna

Block-wisely Supervised Neural Architecture Search with Knowledge Distillation (CVPR 2020)

Stars: ✭ 147 (-18.33%)

Mutual labels: neural-architecture-search

Graphnas

This directory contains code necessary to run the GraphNAS algorithm.

Stars: ✭ 104 (-42.22%)

Mutual labels: neural-architecture-search

Awesome Autodl

A curated list of automated deep learning (including neural architecture search and hyper-parameter optimization) resources.

Stars: ✭ 1,819 (+910.56%)

Mutual labels: neural-architecture-search

Nsga Net

NSGA-Net, a Neural Architecture Search Algorithm

Stars: ✭ 171 (-5%)

Mutual labels: neural-architecture-search

Aw nas

aw_nas: A Modularized and Extensible NAS Framework

Stars: ✭ 152 (-15.56%)

Mutual labels: neural-architecture-search

Sgas

SGAS: Sequential Greedy Architecture Search (CVPR'2020) https://www.deepgcns.org/auto/sgas

Stars: ✭ 137 (-23.89%)

Mutual labels: neural-architecture-search

View All Similar Projects ➔

Lycoris is a lightweight and easy-to-use deep learning framework with neural architecture search.

Lycoris aims to provide developers with an automated service for end-to-end neural network architecture search, enabling developers to obtain better models with fewer hyperparameter configurations. For more detailed usage of Lycoris, please refer to related documents.

At this stage, Lycoris enables you to deploy computation to one or more CPUs in most operating systems. In the future, GPUs will be added to the minimalist architecture.

Features

Lightweight and portable to smart devices.
Support online learning.
Automatic neural network architecture search.
Support for Python. In the near future, other programming languages such as Java, C #, Go, and Rust will be supported.
Cloud-friendly. Based on C++11, it supports most operating systems and compilers.

Installation

C++ version:

cd Lycoris
cmake .
sudo make install

Python bindings:

The Python bindings of the code are based on Pybind11.

git clone "https://github.com/pybind/pybind11.git"
cd pybind11
mkdir build
cd build
cmake ..
make install

(If pybind11 and its header files are already installed, you can ignore the above steps.)

pip install LycorisNet

It can also be obtained via manual compilation:

cd Lycoris/python
cmake .
make

Documents

The following is the documentation for the C ++ version while that of the Python version can be viewed here.

The APIs provided by Lycoris (namespace LycorisNet):

Function	Description	Inputs	Returns
Lycoris(uint32_t capacity, uint32_t inputDim, uint32_t outputDim, const std::string &mode);	Constructor. The class Lycoris is the highest level abstraction of LycorisNet.	capacity: Capacity of Lycoris. inputDim: Input dimension. outputDim: Output dimension. mode: Mode of Lycoris (classify or predict).	An object of the class Lycoris.
~Lycoris();	Destructor.
void preheat(uint32_t num_of_nodes, uint32_t num_of_connections, uint32_t depth);	Preheating process of the neural network cluster.	num_of_nodes: The number of hidden nodes added for each neural network. num_of_connections: The number of connections added for each neural network. depth: Total layers of each neural network.
void evolve(std::vector<std::vector<float> > &input, std::vector<std::vector<float> > &desire);	Evolve the neural network cluster.	input: Input data. desire: Expected output data.
void fit(std::vector<std::vector<float> > &input, std::vector<std::vector<float> > &desire);	Fit all neural networks in the neural network cluster.	input: Input data. desire: Expected output data.
void enrich();	Keep only the best one in the neural network cluster.
std::vector<float> compute(std::vector<float> &input);	Forward Computing of the best individual.	input: Input data.	Returns the output data.
std::vector<std::vector<float> > computeBatch(std::vector<std::vector<float> > &input);	Parallel forward Computing of the best individual.	input: Input data (two dimensions).	Returns the output data (two dimensions).
void resize(uint32_t capacity);	Resize the capacity of the neural network cluster.	As literally.
void openMemLimit(uint32_t size);	Turn on memory-limit.	As literally.
void closeMemLimit();	Turn off memory-limit.
void saveModel(const std::string &path);	Export the current trained model.	path: File path of the current trained model.
void setMutateArgs(std::vector<float> &p);	Set p1 to p4 in the class Args. Parameters are passed in as `std::vector`.	p1: Probability of adding the new node between a connection. p2: Probability of deleting a node. p3: Probability of adding a new connection between two nodes. p4: Probability of deleting a connection.
void setMutateOdds(float odds);	Set the odds of mutating.	The param "odds" means one individual mutates odds times to form odds + 1 individuals.
void setCpuCores(uint32_t num);	Set the number of worker threads to train the model.	As literally.
void setLR(float lr);	Set the learning rate.	As literally.
uint32_t getSize();			Returns the size of the best individual.
uint32_t getInputDim();			Returns the input dimension.
uint32_t getOutputDim();			Returns the output dimension.
uint32_t getCapacity();			Returns capacity of Lycoris.
float getLoss();			Returns the loss.
std::string getMode();			Returns mode of Lycoris (classify or predict).
std::vector<uint32_t> getLayers();			Returns the number of nodes in each layer of the neural network.
std::vector<float> getHiddenLayer(uint32_t pos);	The parameter pos starts at index 0.	pos: The number of the layer needed.	Returns a vector of nodes in a specific layer of the best individual.
static std::string version();			Returns version information and copyright information.

The funtion used to import the pre-trained model (namespace LycorisNet):

Function	Description	Inputs	Returns
Lycoris *loadModel(const std::string &path, uint32_t capacity);	Import the pre-trained model.	path: File path of the pre-trained model. capacity: Capacity of the neural network cluster.	Returns a pointer to the object of class Lycoris.
Lycoris *loadViaString(const std::string &model, uint32_t capacity);	Import the pre-trained model via string.	model: The pre-trained model in the form of string. capacity: Capacity of the neural network cluster.	Returns a pointer to the object of class Lycoris.

More detailed documentation will be released in the form of sample code.

Examples

LycorisAD: an elegant outlier detection algorithm framework based on AutoEncoder.
LycorisR: a lightweight recommendation algorithm framework based on LycorisNet.
LycorisQ: a neat reinforcement learning framework based on LycorisNet.
More examples will be released in the future.

License

Lycoris is released under the LGPL-3.0 license. By using, distributing, or contributing to this project, you agree to the terms and conditions of this license.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].

Stars: ✭ 180

Visit Git Page 🔗Visit User Page 🔗Visit Issues Page (1) 🔗