opendilab/treevalue

TreeValue is a generalized tree-based data structure mainly developed by OpenDILab Contributors.

Almost all the operations can be supported in the form of trees in a convenient way to simplify the structure processing when the calculation is tree-based.

Outline

• Overview
• Getting Started
  • Prerequisite
  • Installation
  • Quick Usage
  • Tutorials
  • External
• Speed Performance
• Change Log
• Feedback and Contribute
• Citation
• License

Overview

When we build a complex nested structure, we need to model it as a tree structure, and the native list and dict in Python are often used to solve this problem. However, it takes a lot of codes and some complex and non-intuitive calculation logic, which is not easy to modify and extend related code and data, and parallelization is impossible.

Therefore, we need a kind of more proper data container, named TreeValue. It is designed for solving the following problems:

• Ease of Use: When the existing operations are applied to tree structures such as dict, they will become completely unrecognizable, with really low readability and maintainability.
• Diversity of Data: In the tree structure operation, various abnormal conditions (structure mismatch, missing key-value, type mismatch, etc.) occur from time to time, and the code will be more complicated if it needs to be handled properly.
• Scalability and Parallelization: When any multivariate operation is performed, the calculation logic needs to be redesigned under the native Python code implementation, and the processing will be more complicated and confusing, and the code quality is difficult to control.

Getting Started

Prerequisite

treevalue has been fully tested in the Linux, macOS and Windows environments and with multiple Python versions, and it works properly on all these platforms.

However, treevalue currently does not support PyPy, so just pay attention to this when using it.

Installation

You can simply install it with pip command line from the official PyPI site.

pip install treevalue

Or just from the source code on github

pip install git+https://github.com/opendilab/treevalue.git@main

For more information about installation, you can refer to the installation guide.

After this, you can check if the installation is processed properly with the following code

from treevalue import __version__
print('TreeValue version is', __version__)

Quick Usage

You can easily create a tree value object based on FastTreeValue.

from treevalue import FastTreeValue

if __name__ == '__main__':
    t = FastTreeValue({
        'a': 1,
        'b': 2.3,
        'x': {
            'c': 'str',
            'd': [1, 2, None],
            'e': b'bytes',
        }
    })
    print(t)

The result should be

<FastTreeValue 0x7f6c7df00160 keys: ['a', 'b', 'x']>
├── 'a' --> 1
├── 'b' --> 2.3
└── 'x' --> <FastTreeValue 0x7f6c81150860 keys: ['c', 'd', 'e']>
    ├── 'c' --> 'str'
    ├── 'd' --> [1, 2, None]
    └── 'e' --> b'bytes'

And t is structure should be like this

Not only a visible tree structure, but abundant operation supports is provided. You can just put objects (such as torch.Tensor, or any other types) here and just call their methods, like this

import torch

from treevalue import FastTreeValue

t = FastTreeValue({
    'a': torch.rand(2, 5),
    'x': {
        'c': torch.rand(3, 4),
    }
})

print(t)
# <FastTreeValue 0x7f8c069346a0>
# ├── a --> tensor([[0.3606, 0.2583, 0.3843, 0.8611, 0.5130],
# │                 [0.0717, 0.1370, 0.1724, 0.7627, 0.7871]])
# └── x --> <FastTreeValue 0x7f8ba6130f40>
#     └── c --> tensor([[0.2320, 0.6050, 0.6844, 0.3609],
#                       [0.0084, 0.0816, 0.8740, 0.3773],
#                       [0.6523, 0.4417, 0.6413, 0.8965]])

print(t.shape)  # property access
# <FastTreeValue 0x7f8c06934ac0>
# ├── a --> torch.Size([2, 5])
# └── x --> <FastTreeValue 0x7f8c069346d0>
#     └── c --> torch.Size([3, 4])
print(t.sin())  # method call
# <FastTreeValue 0x7f8c06934b80>
# ├── a --> tensor([[0.3528, 0.2555, 0.3749, 0.7586, 0.4908],
# │                 [0.0716, 0.1365, 0.1715, 0.6909, 0.7083]])
# └── x --> <FastTreeValue 0x7f8c06934b20>
#     └── c --> tensor([[0.2300, 0.5688, 0.6322, 0.3531],
#                       [0.0084, 0.0816, 0.7669, 0.3684],
#                       [0.6070, 0.4275, 0.5982, 0.7812]])
print(t.reshape((2, -1)))  # method with arguments
# <FastTreeValue 0x7f8c06934b80>
# ├── a --> tensor([[0.3606, 0.2583, 0.3843, 0.8611, 0.5130],
# │                 [0.0717, 0.1370, 0.1724, 0.7627, 0.7871]])
# └── x --> <FastTreeValue 0x7f8c06934b20>
#     └── c --> tensor([[0.2320, 0.6050, 0.6844, 0.3609, 0.0084, 0.0816],
#                       [0.8740, 0.3773, 0.6523, 0.4417, 0.6413, 0.8965]])
print(t[:, 1:-1])  # index operator
# <FastTreeValue 0x7f8ba5c8eca0>
# ├── a --> tensor([[0.2583, 0.3843, 0.8611],
# │                 [0.1370, 0.1724, 0.7627]])
# └── x --> <FastTreeValue 0x7f8ba5c8ebe0>
#     └── c --> tensor([[0.6050, 0.6844],
#                       [0.0816, 0.8740],
#                       [0.4417, 0.6413]])
print(1 + (t - 0.8) ** 2 * 1.5)  # math operators
# <FastTreeValue 0x7fdfa5836b80>
# ├── a --> tensor([[1.6076, 1.0048, 1.0541, 1.3524, 1.0015],
# │                 [1.0413, 1.8352, 1.2328, 1.7904, 1.0088]])
# └── x --> <FastTreeValue 0x7fdfa5836880>
#     └── c --> tensor([[1.1550, 1.0963, 1.3555, 1.2030],
#                       [1.0575, 1.4045, 1.0041, 1.0638],
#                       [1.0782, 1.0037, 1.5075, 1.0658]])

Tutorials

For more examples, explanations and further usages, take a look at:

• Quick Start
  • Create a Simplest Tree
  • Create a Slightly Complex Tree
• Basic Usage
  • Create a Tree
  • Edit a Tree
  • Do Index or Slice Calculation on The Tree
  • Do Math Calculation on The Tree
  • Make Function Tree-Supported
• Advanced Usage
  • Function Modes
  • Inheriting on Trees
  • Process Missing Values
  • Functional Utilities
  • Structural Utilities
  • Tree Utilities
  • Flatten Utilities
  • IO Utilities
  • Object Oriented Usage
  • Costumize My TreeValue Class
  • Visualization of TreeValue

External

We provide an official treevalue-based-wrapper for numpy and torch called DI-treetensor since the treevalue is often used with libraries like numpy and torch. It will actually be helpful while working with AI fields.

Speed Performance

Here is the speed performance of all the operations in FastTreeValue; the following table is the performance comparison result with dm-tree. (In DM-Tree, the unflatten operation is different from that in TreeValue, see: Comparison Between TreeValue and DM-Tree for more details.)

The following 2 tables are the performance comparison result with jax pytree.

This is the comparison between dm-tree, jax-libtree and us, with flatten and mapping operations (lower value means less time cost and runs faster)

The following table is the performance comparison result with tianshou Batch.

And this is the comparison between Tianshou Batch and us, with cat , stack and split operations (lower value means less time cost and runs faster)

Test benchmark code can be found here:

• Comparison with dm-tree
• Comparison with jax-libtree
• Comparison with tianshou Batch

Change Log

Feedback and Contribute

Welcome to OpenDILab community - treevalue!

If you meet some problem or have some brilliant ideas, you can file an issue.

Scan the QR code and add us on Wechat:

Or just contact us with slack or email ([email protected]).

Please check Contributing Guidances.

Thanks to the following contributors!

Citation

@misc{treevalue,
    title={{TreeValue} - Tree-Structure Computing Solution},
    author={TreeValue Contributors},
    publisher = {GitHub},
    howpublished = {\url{https://github.com/opendilab/treevalue}},
    year={2021},
}

License

treevalue released under the Apache 2.0 license. See the LICENSE file for details.