Stamox

Hello Stamox, Why Another Wheel?

Why Another Wheel?

What stamox does is really simple, just make it possible, it aims to provide a statistic interface for JAX. But nowadays, we have so many statistic packages around the world varying from languages, for python, statsmodels just works great, for R, tidyverse derived packages are so delicate and easy to use. So why build another wheel?

Three reasons I think:

• Personal interest, as a student of statistics, I want to learn more about statistics and machine learning, proficient knowledge comes from books but more from practice, write down the code behind the theory is a good way to learn.

• Speed, JAX is really fast, and Equinox is a good tool to make JAX more convenient, backend of JAX is XLA, which makes it possible to compile the code to GPU or TPU, and it is really fast.

• Easy of Use, %>% is delicate operation in R, it combines the functions to a pipe and make the code more readable, and stamox is inspired by it, and I want to take a try to make it convenient in python with >>.

And here're few benchmarks:

generate random variables

calculate cdf

Installation

pip install -U stamox
# or
pip install git+https://github.com/JiaYaobo/stamox.git

Documentation

More comprehensive introduction and examples can be found in the documentation.

Quick Start

Similar but faster distribution functions to R

You can simply import all functions from stamox.functions

from stamox.functions import dnorm, pnorm, qnorm, rnorm
import jax.random as jrandom

key = jrandom.PRNGKey(20010813)

# random
x = rnorm(key, sample_shape=(1000, ))
# cdf
prob = pnorm(x)
# ppf
qntl = qnorm(prob)
# pdf
dnorm(x)

Fearless Pipeable

>> is the pipe operator, which is the similar to |> in F# and Elixir or %>% in R.

• You can simply import all pipeable functions from pipe_functions

import jax.random as jrandom
import stamox.pipe_functions as PF
from stamox import pipe_jit

key = jrandom.PRNGKey(20010813)

@pipe_jit
def f(x):
    return [3 * x[:, 0] + 2 * x[:, 1] - x[:, 2], x] # [y, X]
pipe = PF.rnorm(sample_shape=(1000, 3)) >> f >> PF.lm
state = pipe(key)
print(state.params)

Linear Regression with Formula

import pandas as pd
import numpy as np
from stamox.functions import lm # or from stamox.pipe_functions import lm


x = np.random.uniform(size=(1000, 3))
y = 2 * x[:,0] + 3 * x[:,1] + 4 * x[:,2] + np.random.normal(size=1000)
df = pd.DataFrame(x, columns=['x1', 'x2', 'x3'])
df['y'] = y

lm(df, 'y~x1+x2+x3').params

• Custom Functions Pipeable

from stamox import make_pipe, make_partial_pipe, Pipeable
import jax.numpy as jnp
import jax.random as jrandom

x = jnp.ones((1000, ))
# single input, simply add make pipe
@make_pipe
def f(x):
    return x ** 2

# multiple input, decorate with make partial pipe
@make_partial_pipe
def g(x, y):
    return x + y

# x -> f -> g(y=2.) -> f -> g(y=3.) -> f
h = Pipeable(x) >> f >> g(y=2.) >> f >> g(y=3.) >> f
# h is a Pipeable object, you can call it to get the result
print(h())

• Compatible With JAX and Equinox

You can use autograd features from JAX and Equinox with Stamox easily.

import jax.numpy as jnp
from stamox import make_partial_pipe
from equinox import filter_jit, filter_vmap, filter_grad

@make_partial_pipe
@filter_jit
@filter_vmap
@filter_grad
def f(x, y):
    return y * x ** 3

# df/dx = 3y * x^2
g = f(y=3.) # derive with respect to x given y=3
g(jnp.array([1., 2., 3.]))

Or vmap, pmap, jit features integrated with Stamox:

from stamox import pipe_vmap, pipe_jit

@pipe_vmap
@pipe_jit
def f(x):
    return x ** 2

g = f >> f >> f
print(g(jnp.array([1, 2, 3])))

Acceleration Support

JAX can be accelerated by GPU and TPU. So, Stamox is compatible with them.

See More

JAX

Equinox