pytreeclass

Installation |Description |Quick Example |StatefulComputation |Benchamrks |Acknowledgements

🛠️ Installation

pip install pytreeclass

Install development version

pip install git+https://github.com/ASEM000/PyTreeClass

📖 Description

PyTreeClass is a JAX-compatible class builder to create and operate on stateful JAX PyTrees in a performant and intuitive way, by building on familiar concepts found in numpy, dataclasses, and others.

See documentation and 🍳 Common recipes to check if this library is a good fit for your work. If you find the package useful consider giving it a 🌟.

⏩ Quick Example

import jax
import jax.numpy as jnp
import pytreeclass as pytc

@pytc.autoinit
class Tree(pytc.TreeClass):
    a: float = 1.0
    b: tuple[float, float] = (2.0, 3.0)
    c: jax.Array = jnp.array([4.0, 5.0, 6.0])

    def __call__(self, x):
        return self.a + self.b[0] + self.c + x


tree = Tree()
mask = jax.tree_map(lambda x: x > 5, tree)
tree = tree\
       .at["a"].set(100.0)\
       .at["b"].at[0].set(10.0)\
       .at[mask].set(100.0)

print(tree)
# Tree(a=100.0, b=(10.0, 3.0), c=[  4.   5. 100.])

print(pytc.tree_diagram(tree))
# Tree
# ├── .a=100.0
# ├── .b:tuple
# │   ├── [0]=10.0
# │   └── [1]=3.0
# └── .c=f32[3](μ=36.33, σ=45.02, ∈[4.00,100.00])

print(pytc.tree_summary(tree))
# ┌─────┬──────┬─────┬──────┐
# │Name │Type  │Count│Size  │
# ├─────┼──────┼─────┼──────┤
# │.a   │float │1    │      │
# ├─────┼──────┼─────┼──────┤
# │.b[0]│float │1    │      │
# ├─────┼──────┼─────┼──────┤
# │.b[1]│float │1    │      │
# ├─────┼──────┼─────┼──────┤
# │.c   │f32[3]│3    │12.00B│
# ├─────┼──────┼─────┼──────┤
# │Σ    │Tree  │6    │12.00B│
# └─────┴──────┴─────┴──────┘

# ** pass it to jax transformations **
# works with jit, grad, vmap, etc.

@jax.jit
@jax.grad
def sum_tree(tree: Tree, x):
    return sum(tree(x))

print(sum_tree(tree, 1.0))
# Tree(a=3.0, b=(3.0, 0.0), c=[1. 1. 1.])

📜 Stateful computations

Under jax.jit jax requires states to be explicit, this means that for any class instance; variables needs to be separated from the class and be passed explictly. However when using TreeClass no need to separate the instance variables ; instead the whole instance is passed as a state.

Using the following pattern,Updating state functionally can be achieved under jax.jit

import jax
import pytreeclass as pytc

class Counter(pytc.TreeClass):
    def __init__(self, calls: int = 0):
        self.calls = calls

    def increment(self):
        self.calls += 1
counter = Counter() # Counter(calls=0)

Here, we define the update function. Since the increment method mutate the internal state, thus we need to use the functional approach to update the state by using .at. To achieve this we can use .at[method_name].__call__(*args,**kwargs), this functional call will return the value of this call and a new model instance with the update state.

@jax.jit
def update(counter):
    value, new_counter = counter.at["increment"]()
    return new_counter

for i in range(10):
    counter = update(counter)

print(counter.calls) # 10

➕ Benchmarks

Benchmark flatten/unflatten compared to Flax and Equinox

CPU	GPU

Benchmark simple training against `flax` and `equinox`

Training simple sequential linear benchmark against flax and equinox

Num of layers	Flax/PyTC time	Equinox/PyTC time
10	1.427	6.671
100	1.1130	2.714

📙 Acknowledgements

Lenses
Treex, Equinox, tree-math, Flax PyTreeNode, TensorFlow, PyTorch
Lovely JAX

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pytreeclass-0.6.0.tar.gz (56.1 kB view details)

Uploaded Jul 31, 2023 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

pytreeclass-0.6.0-py3-none-any.whl (46.1 kB view details)

Uploaded Jul 31, 2023 Python 3

File details

Details for the file pytreeclass-0.6.0.tar.gz.

File metadata

Download URL: pytreeclass-0.6.0.tar.gz
Upload date: Jul 31, 2023
Size: 56.1 kB
Tags: Source
Uploaded using Trusted Publishing? No
Uploaded via: twine/4.0.2 CPython/3.11.4

File hashes

Hashes for pytreeclass-0.6.0.tar.gz
Algorithm	Hash digest
SHA256	`9f35429dad68ae694e4d5b9e98f835eab676e10898a302d273095e0668183974`
MD5	`b220f8b50b22ffd7db7786f67b7aef33`
BLAKE2b-256	`f9305cba8759b6cf1f8d7ad1c8286de847e40f243ec339617afd5bb1e11d87ae`

See more details on using hashes here.

File details

Details for the file pytreeclass-0.6.0-py3-none-any.whl.

File metadata

Download URL: pytreeclass-0.6.0-py3-none-any.whl
Upload date: Jul 31, 2023
Size: 46.1 kB
Tags: Python 3
Uploaded using Trusted Publishing? No
Uploaded via: twine/4.0.2 CPython/3.11.4

File hashes

Hashes for pytreeclass-0.6.0-py3-none-any.whl
Algorithm	Hash digest
SHA256	`f3a2a7cd9d98a93f4ce0b10e1bf33904ccaf437232eea32aeb802ffeef34c785`
MD5	`593eb37fd259ab386ed513e269bd5b81`
BLAKE2b-256	`e29c2f7a83fbc5eb26692d4e279fa585d62f2eed1b85db2df61fa5f9ed7f5bee`