How to use decorators

Learn about the different ways to use compilation and transpilation functions.

⚠️ If you are running this notebook in Colab, you will have to install Ivy and some dependencies manually. You can do so by running the cell below ⬇️

If you want to run the notebook locally but don’t have Ivy installed just yet, you can check out the Get Started section of the docs.

!pip install ivy

Unify

Firstly, let’s create the dummy numpy arrays as before:

# import numpy
import numpy as np

# create random numpy arrays for testing
x = np.random.uniform(size=10)
mean = np.mean(x)
std = np.std(x)

Let’s assume that our target framework is tensorflow:

import ivy
import tensorflow as tf
ivy.set_backend("tensorflow")

x = tf.constant(x)

In the example below, the ivy.unify function is called as a decorator.

import torch

@ivy.unify(source="torch")
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # unification happens here

ivy.array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])

The function can still be called either eagerly or lazily when calling as a decorator. The example above is lazy, whereas the example below is eager:

@ivy.unify(source="torch", args=(x,))
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # already unified

ivy.array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])

Compile

In the example below, the ivy.compile function is also called as a decorator. (Note that this is now an Ivy function!)

@ivy.compile
def normalize(x):
    mean = ivy.mean(x)
    std = ivy.std(x, correction=1)
    return ivy.divide(ivy.subtract(x, mean), std)

normalize(x) # compilation happens here

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])>

Likewise, the function can still be called either eagerly or lazily when calling as a decorator. The example above is lazy, whereas the example below is eager:

@ivy.compile(args=(x,))
def normalize(x):
    mean = ivy.mean(x)
    std = ivy.std(x, correction=1)
    return ivy.divide(ivy.subtract(x, mean), std)

normalize(x) # already compiled

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])>

Transpile

In the example below, the ivy.transpile function is called as a decorator.

@ivy.transpile(source="torch", to="tensorflow")
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # transpilation happens here

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])>

The function can still be called either eagerly or lazily when calling as a decorator. The example above is lazy, whereas the example below is eager:

@ivy.transpile(source="torch", to="tensorflow", args=(x,))
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # already transpiled

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-1.09422972, -0.46009917,  1.0881108 ,  1.86487021,  0.83629996,
       -1.10654466, -0.89883457,  0.02893805,  0.15644584, -0.41495672])>

Round Up

That’s it, you now know how ivy.unify, ivy.compile and ivy.transpile can all be used as function decorators! Next, we’ll start exploring the transpilation of more involved objects, beginning with libraries 📚