Ray Tests 🧪

Ray has potential to distrupt a lot of services. It allows for distrubted compute with just a few changes in the python code. Let's explore the API provided by Ray to improve performance on multi-core or multi-node operations.

Ray Github

Ray Documentation

# Install Ray using pip !pip install ray

import ray import time import os

ray.__version__

Toy Example 🧸

# A toy example to demonstrate normal function in python which might let's say 2 seconds to complete def basic_function(x): time.sleep(2) return x*x

# Exact replicate of the previous function but with a decorator of ray @ray.remote def ray_function(x): time.sleep(2) return x*x

%%time # Run the basic python function 4 times y = [basic_function(x) for x in range(4)] y

# Init function of ray the resources can be specified here, for example num_cpus, num_gpus etc. ray.init(num_cpus=os.cpu_count())

%%time # Ray function can be used with same arguments but need remote() is added futures = [ray_function.remote(x) for x in range(4)]

%%time # Get function takes all the list of futures from the previous output and materalize the results y_h = ray.get(futures) y_h

# Check if ray output y_h and python outout y are the same indeed assert y_h == y

Conslusion 📝

We saw a simple toy example of how to use Ray on a single node, Ray can be scaled to multiple nodes and handle multi-core just as easily as the above example. For more information about all the cool supported libaries built on top of Ray refer to the official documentation (link of top). It's really exciting to see Ray work and still keep it as simple as possible. Happy Exploring!! 🤓

Ray Tests 🧪

Toy Example 🧸

Conslusion 📝

Recommended on Deepnote

Stock Market Analysis

The 10 Best Ways to Create NumPy Arrays

Wide Residual Networks