Pere Ferrera Bertran

Father, data / software / ML / AI engineer, musician, existentialist

[Pere Ferrera Bertran @ferrerabertran]

AI coding: perhaps the hottest topic in software engineering right now - here’s a thoughtful take on how to survive the hype - https://t.co/pesN12RqLX

[Pere Ferrera Bertran @ferrerabertran]

I just published "A case of GIL starvation: my toughest software engineering investigation" https://t.co/DRICTFzR89

[Pere Ferrera Bertran @ferrerabertran]

@nathanmarz Will there also be a huge gain in using the platform for applications that do not deal with large amounts of data or is it mainly thought for big data?

[Pere Ferrera Bertran @ferrerabertran]

I wanted to have a feeling on what it is like to interact with a LLM for coding without involving any IDE at all. All communication done through github issues https://t.co/cDJV3ntEMT #GPT #AI #LLM #LLMs

[Pere Ferrera Bertran @ferrerabertran]

Towards accountable LLMs: making GPT-3 reconsider its answers #ai #gpt3 #chatgpt #llms
https://t.co/nl7pdvntCW

[Pere Ferrera Bertran @ferrerabertran]

Towards better automated testing (co-authored with ChatGPT): a post about mutation testing, property-based testing, ... https://t.co/SE4t2E36xL

[Pere Ferrera Bertran @ferrerabertran]

All you ever wanted to know about multi-threading & multi-processing in python (an interactive notebook) #python #programming #concurrency #multithreading #multiprocessing https://t.co/9dPIHSGd0N

[Pere Ferrera Bertran @ferrerabertran]

@formerlyXurde En Zehlendorf desde 2018... Para mí, Steglitz/Friedenau es como el centro de la ciudad 😅

[Pere Ferrera Bertran @ferrerabertran]

@formerlyXurde Friedrich-Wilhelm-Platz! 👍

[Pere Ferrera Bertran @ferrerabertran]

One year participating in the Numerai Machine Learning competition: 7 lessons learned - https://t.co/1xJRb1mjqg @numerai #numerai #machinelearning

[Numerai @numerai]

Numerai outperforms market neutral hedge funds by 29% and attracts over $150m in institutional capital
https://t.co/Gs8yw3pdQO

[Pere Ferrera Bertran @ferrerabertran]

@sturlese @Alex_E_Benton @numerai Doncs sí, al final he pogut re-entrenar els models amb Parquet int8 i fent un ensemble de 4 sense era overlapping. Per nous experiments hauria de probar Colab Pro! Gràcies per la recomanació!

[Pere Ferrera Bertran @ferrerabertran]

@Prutyke @numerai How many attempts did it take?

[Pere Ferrera Bertran @ferrerabertran]

@anacoding Is this Spearman on the validation dataset? My models have a similar score there

[Pere Ferrera Bertran @ferrerabertran]

"A transformer based architecture for predicting if your baby is ready to fall asleep within 5 minutes if attempting to take a nap, achieving superior SOTA performance" - P.Ferrera et. al, 2021

[Pere Ferrera Bertran @ferrerabertran]

@Alex_E_Benton Similar thing for me with my gboost model. However my MLP trained in a similar fashion is in green for these rounds. It apparently has less feature exposure out of the box...

[Pere Ferrera Bertran @ferrerabertran]

@Abe_6006 @numerai 1 x gradient boosting and 1 x MLP. For now. Planning to try more models / ideas.

[Pere Ferrera Bertran @ferrerabertran]

@sturlese @numerai Pytorch! Love it

[Pere Ferrera Bertran @ferrerabertran]

First DL model outperformed my gboost model which is currently in red. Might want to start staking on it... Almost same scores in cross validation and test dataset, yet such a different behavior! @numerai #numerai #DeepLearning #MachineLearning

[Pere Ferrera Bertran @ferrerabertran]

@Mandelliant @numerai Nice. How did you cross validate, based on eras or on random splits?