Skip to content

Núria Puig

@nuria_data

Data scientist a Barcelona 📊 Python, pandas, una mica de ML. Catalan, Spanish, English, whatever works. Sometimes I post in català, sorry not sorry.

  • Barcelona, Spain
  • Catalan · Spanish · English
  • Joined Jul 2026
  • Last posted Oct 13, 2026

Known for

Tags

Posts
4
Followers
11
Following
10
Reputation
150

Posts

Everything this author has published.

Núria Puig233
t/stochastic-parrots

Avui he descobert que el 12% de les ressenyes del nostre dataset d'entrenament les ha escrit un LLM. Totes comencen igual: "Com a client satisfet...". Bueno. Ara entrenem un model amb text d'un altre model. Què pot sortir malament? 🙃

5

  • Dana Kowalski
    Read it with a translator. Worth a full post when you have numbers: how much did removing them change your model's outputs? I suspect the effect on style is larger than on accuracy.
    1 reply
Núria Puig468
t/my-model-is-fine

Pandera, Great Expectations or plain asserts for data validation in a small team?

Hola! Small data team (3 people), Python, pandas, some dbt, Airflow. We have zero data validation, and last month a supplier changed a column from euros to cents without telling anybody. Our revenue dashboard showed 100x for two days 🙃 Nobody noticed until the CFO was very happy, and then very angry. I want to add validation at the ingestion step. The options I see: pandera : schemas in Python, looks light. Great…

6

  • Accepted answer

    Lin Wei
    Small team, I'd do two layers. 1. pandera at ingestion for schema and ranges. It's just Python, it reads like documentation and it fails loudly: import pandera as pa schema = pa.DataFrameSchema({ 'amount_eur': pa.Column(float, pa.Check.in_range(0, 50_000)), 'currency': pa.Column(str, pa.Check.isin(['EUR'])), }) schema.validate(df, lazy=True) 2. dbt tests (not_null, unique, accepted_values, plus a custom one) on the models you already have. For euros to cents specifically, a range check won't catch it if the max is generous. What catches it is a distribution check against history: today's median amount vs the last 30 days. If it moves 10x, fail. That's a 15-line custom test in dbt. GX can do this too, but for 3 people the setup cost is real.
    1 reply
Núria Puig343
t/ai_hype_check2 min

Antes de culpar al modelo, mira tus etiquetas

Lo prometí en el otro post, así que aquí va en castellano (y algo de català, perdón). Teníamos un clasificador de reseñas de clientes con 5 categorías: producto, envío, precio, atención al cliente y "otro". Llevábamos dos meses probando modelos: TF-IDF con regresión logística, un BERT multilingüe afinado y un LLM con few-shot. Los tres se quedaban en 0.78-0.80 de F1. Siempre lo mismo. Sospechoso, ¿no? El experimento…

7

  • Javier Montero
    Vaya tela con la empresa externa. Déjame adivinar: la contratasteis por precio y etiquetaban becarios a destajo. Clásico de cárnica. Buen post, Núria, aunque seas del Barça.
    1 reply
Núria Puig1.2K
t/ai_hype_check3 min

Your model didn't get better, your test set leaked (a checklist)

A ver. Three times this year a model "improved" by a lot, and three times it was leakage. The last one was a fine-tuned classifier that jumped from 0.81 to 0.93 F1, and my manager already had a slide 🙃 This is the checklist I run now before I believe anything. 1. Exact and near duplicates between train and test The 80% of my leakage problems are this. Customer reviews, support tickets, product descriptions: people…

7

  • Sofía Ramírez
    no soy de datos pero lo entendí todo, gracias Núria 💜 ¿lo vas a traducir para codigo_en_espanol?
    1 reply

Best answers

Get the best of Dynamo Geeks weekly

One email each Monday with the most useful posts. No spam; unsubscribe in one click.