marquetry 0.0.1

Simple Machine Learning Framework

Marquetry

Marquetry means Yosegi-zaiku, a traditional Japanese woodworking technique, in Japan.
It is a beautiful culture craft originated in Japan, which is a box or ornament or so by small wooden pieces.
The design is UNIQUE, it depends on the arrangement of the wood tips.
I believe Deep Learning is similar with the concept.
Deep Learning models are constructed through the combination of the layers or functions. Just as a slight variation in arrangement can result in a completely distinct model.
I want you can enjoy the deep/machine learning journey like you craft a Marquetry from combination of various wood tips.

About this Framework

You can use this framework for help your learning Machine/Deep Learning.
This framework is written only Python, so you can understand the implementation easily if you are python engineer.
For simplify the construct, there are un-efficiency implementation.
I develop this framework to enjoy learning the construction of the machine/machine learning not Practical Usage.
I hope to enjoy your journey!

Directory

├── README.md
├── marquetry
│   ├── __init__.py
│   ├── pre_implementation ... preview implement components
│   │   └── svm.py
│   ├── conventional_ml ... conventional machine learning components
│   │   └── tree.py ... DecisionTree and RandomForest
│   │
│   ├── core.py ... Core components of the marquetry
│   ├── datasets.py ... Dataset like "MNIST"/"Titanic" and so
│   ├── dataloaders.py ... Dataloader components
│   ├── functions.py ... Functions for layer/model construction
│   ├── cuda_backend.py ... using CuPy component and transform NumPy and CuPy component
│   ├── layers.py ... Layers conponents
│   ├── models.py ... Example models
│   ├── optimizers.py ... Model optimizer
│   ├── preprocess.py ... Preprocess Script
│   ├── transformers.py ... Data transformers
│   └── utils.py ... other utils
├── setup.py
└── tests

Dependencies

You need to fill the below version requirement and import external libraries.

• Python 3 | 3.8 or later
• NumPy
• Pandas

Optional

• CyPy

for display the calculation graph

• Pillow

for test script

• PyTorch

License

This project is licensed under the MIT License.

Reference Source

This framework started to be developed based on dezero.
Originally, the dezero was developed based on Chainer (and PyTorch) architecture.
Therefore, there are much similar architecture between dezero(Chainer) and this like the algorithm of the auto-gradient and so.
If you want to know about this framework deeply, I suggest to visit the dezero and Chainer repository. (PyTorch is more complex but beautiful.)

And, I respect the dezero author and his books are very curiously and easy to understandable.
If you want to start journey for deep learning world, I suggest to read his books.