Engaging ML Tracking and Visualization : Sit Back and Relax

Paras Koundal
Analytics Vidhya
Published in
4 min readJul 25, 2021

--

Chapter 1 — W&B

Learn Something New: The Golden Pheasant are popularly known as the ‘Chinese Pheasant’ as they are native to Western and Central China, living primarily in the mountainous forests. From The New York Public Library — https://on.nypl.org/3zwIhFJ

The world of Machine and Deep Learning (ML & DL) is growing at a fantastic pace. As presented in the “The AI Index 2021 Annual Report” and shown in the following graph, in the year 2020, US researchers alone submitted more than 11k AI-Related publications on arXiv. Combining the numbers from other places like the EU and China, AI and its subcomponents ML & DL are among the most promising and emerging fields in computer science, with wide-ranging applications in other areas like Biology, Physics, and Chemistry.

Geographic Distribution of AI-Related publications on arXiv from 2015–2020 (Data Source: arXiv, 2020; Citation: Daniel Zhang, Saurabh Mishra, Erik Brynjolfsson, et. al. , “The AI Index 2021 Annual Report,” AI Index Steering Committee, Human-Centered AI Institute, Stanford University, Stanford, CA, March 2021. https://stanford.io/3i0PEz4 )

Hidden behind all this promise lies the hard work of thousands of AI practitioners. Putting an AI model into production is preceded by a lot of time-intensive and computation-intensive steps, like data collection, data preprocessing, train multiple iterations of the model, optimizing the model, and analyzing the model on untrained data. Successful training of a big model, with millions of trainable and tunable parameters, might take hundreds (sometimes even thousands or more) of iterations. As models get bigger ; while ML/DL computations get faster and we train even more models by every passing day; manually tracking, visualizing, and recording meta-info becomes non-viable. Lucky for us, amazing people out there are building a lot of valuable and beautiful tools which automate the tracking, visualization and recording part for us. So, now we can pay our full attention to the vital task of inventing new algorithms and optimizing our models. Here I’ll discuss one such tool namely “Weights and Biases ( https://wandb.ai/ )”.

Weights and Biases

So what can W&B do for you: As they claim and is tested by me, in comparison to other tools it excels at a lot of fronts. They are:

  1. Ease of Integration
    With as few as 4–5 additional lines into your current codebase, you can…

--

--

Paras Koundal
Analytics Vidhya

Hi there. 👋🏾I am a Particle Physicist working in the in the field of Cosmic-Ray Physics at IceCube Neutrino Observatory in Antarctica. Visit paraskoundal.com