Marçal Serrate

Director of Data Technology @WeAreAzerion - #AdTech - Advanced Analytics - Big Data & Machine Learning

[interstellarninja @intrstllrninja]

we've got a 1M context window open-source llm from a shanghai lab out in the wild and you're still stuck with your rag pipeline? @intern_lm

[Marçal Serrate @mserrate]

Check out the current state of #attribution proposals and how Hybrid Theory measures #marketing effectiveness without leveraging inconsistent models such as "last-click attribution"

[Marçal Serrate @mserrate]

Glad you like it @chrisbond63!

[Chris Bond @chrisbond63]

Levels of #Burnout are especially present for #IT professionals who are always making themselves available. Leaders can help by making work more meaningful and connecting with employees on a personal level. Thanks, @mserrate! @4enterprisers
https://t.co/o6lfaQQJoF

[Marçal Serrate @mserrate]

I really hope that while moving away from 3P cookies we are able to measure what it really matters & not just the last click activity.
Otherwise we will need to rely on techniques such as Marketing Mix Modelling and conversion lift to effectively measure marketing investments

[Marçal Serrate @mserrate]

Current state of measurement proposals without 3P cookies in #AdTech:
📈Event-Level Reporting: Associate an ad click with conversion data.
• Safari (Apple) proposed Private-Click-Measurement
• Chrome (Google) has proposed “Event-Level Reports” in their Attribution Reporting API

[Marçal Serrate @mserrate]

All of them comes with different trade-offs between privacy & utility. Finding the right balance is what makes it difficult to reach consensus.

Most of them support last touch attribution only, and as we know, aside from not being accurate it generates incentive problems

[Python Barcelona @PyBCN]

"Burnout Paradise: buenas prácticas para evitar el burnout en desarrollo", by Alejandro Nicolás (@sander_pm) from Hybrid Theory (@HybridTheory_GL)

[Marçal Serrate @mserrate]

Find out what we’re doing at @HybridTheory_GL to support the #cookieless future and one of our tests around #FLoC

[Matei Zaharia @matei_zaharia]

#ApacheSpark 3.0 greatly simplifies writing Python user-defined functions through type hints, and makes it easier for your functions to process data efficiently in batches via Pandas and Apache Arrow. Check out how to use them: https://t.co/f72BhSpJl6

[Marçal Serrate @mserrate]

Great piece of work from our analytics team!

[Netflix Research @NetflixResearch]

We are very excited to announce that we have open-sourced Metaflow, our human-friendly Python library for Data Science! We worked together with @awscloud to ensure that the open-source release is easily usable outside @Netflix. Learn more at https://t.co/cSyzVlA9ny

[World of Engineering @engineers_feed]

Can you solve it?

[Marçal Serrate @mserrate]

@Marc_Rubino @Typs_es Sort Marc!!

[Petter Törnberg @pettertornberg]

All these datasets have the same statistical properties.
Including the dinosaur.

A reminder to be careful with means, variances and correlations - they can be deceptive.

[Marçal Serrate @mserrate]

#TensorFlow Probability looks really promising for bayesian inference https://t.co/nk7ORJFfjt

[Jonathan Libov @libovness]

Remember: Every time you misidentify a traffic light in a Captcha, a self-driving car hits a pedestrian

[Marçal Serrate @mserrate]

Retail forecasting example: https://t.co/WmNs8Roaqq

[Jordi Aranda @jordi_aranda]

TIL how you partition your data REALLY matters: boost in performance in production job (from 4 hours to 1, yeah!) via @mserrate

[Matei Zaharia @matei_zaharia]

Speed up data processing by querying data before parsing it with Sparser, a new technique from the DAWN group at Stanford: https://t.co/CB2bFbKQE4