Step Zero for Data Scientist

Enabling Multilingual Programming // Rodolfo Nunes

On this podcast, we had Rodolfo Nunes, Senior MLOps Engineer at Entel. He talked about why data scientists should know how to code properly.

Building software is the final objective for what a data scientist works on, although it is a very specific kind of software. It's software that takes data, transforms data, and outputs a model or a prediction, but at the end of the day, you want to get a product out of it.

Therefore, it's essential that a data scientist knows basic software practices like version control. Having a proper structure makes code easy to understand and use even without comments.

Video || Spotify || Apple

Etsy's ML Battle Stories

At the previous New York Meetup, Alaa Awad, a Staff Machine Learning Engineer at Etsy, talked about Etsy's ML battle stories.

Model Architecture
Etsy had a baseline model that was an ensemble of logistic regression and gradient-boosted decision trees. Over the years it has been passed around the engineers. It was able to efficiently predict the likelihood of an ad getting clicked by a user.

They introduced a new model architecture that uses Deep Learning. This required revamping the entire ecosystem around the architecture, which includes frameworks, ETL, features e.t.c

However, it degraded in production after a while. After discovering data upstream and feature issues caused the model to degrade, Etsy invested in monitoring upstream data feature development. It constitutes jobs that look at missing percentages of features, and summary statistics. It is also fortified with unit tests and PR checks.

You Tube

Monitoring Machine Learning Applications // Lina Weichbrodt

Lina Weichbrodt is a machine learning consultant with 10+ years of experience developing scalable machine learning models for millions of users and running them in production.

This blog presents an easy-to-implement prioritization approach to use either your own backend monitoring tools or a vendor monitoring tool. It is based on more than 30 large-scale models they have run in production over the last ten years.

Read Here

Understanding Machine Learning Systems by Chip Huyen.

Objective Functions
To learn, an ML model needs an objective function to guide the learning process. An objective function is also called a loss function because the objective of the learning process is usually to minimize (or optimize) the loss caused by wrong predictions.

For supervised ML, this loss can be computed by comparing the model’s outputs with the ground truth labels using a measurement like root mean squared error (RMSE) or cross-entropy.

Choosing an objective function is usually straightforward, though not because objective functions are easy. Coming up with meaningful objective functions requires algebra knowledge, so most ML engineers just use common loss functions like RMSE or MAE (mean absolute error) for regression, logistic loss (also log loss) for binary classification, and cross-entropy for multiclass classification.

Read Here

Two Days of LLMOps Learning // Arize AI

This year’s Arize:Observe is focused on the intersection of large language models, generative AI, and ML observability.

Sessions include:

A Conversation with OpenAI About GPT-4 and Its Implications
Prompt Engineering in the Real World with PromptLayer
How LlamaIndex Brings Your Data to LLMs
A Practical Perspective on Using LLMs with Hugging Face
Training Billion Parameter LLMs with MosaicML
Embeddings at Spotify’s Scale, How Hard Can It Be?
Scale AI on Reinforcement Learning with Human Feedback

Additional speakers hail from AB InBev, Anyscale, Bazaarvoice, Chick-fil-A, Chime, Doordash, Etsy, MIT, Scotiabank, Shopify, Vivun, and over a dozen others.

Peruse the agenda and save your spot for this free, virtual learning experience:

Register