Tomas Bonfiglio

Data Scientist
Hey! I'm a data scientist from Argentina, looking to share and learn about machine learning, mainly in classification or clustering problems :D
Hello! Hola!
  1. I'm Tomas from Buenos Aires, Argentina πŸ‡¦πŸ‡·Β 
  2. I work as a Data Scientist in RockingData, an awesome ML/AI consulting and product startup. I'm particulary interested in classification or clustering problems, explainability and teaching. I use Python and I'm learning some Julia
  3. I think the most exciting project I've worked on has been a resignment prediction engine but It's kinda tied with some interesting clusters of clients πŸ˜„

Like Comment

Using CPU Faiss | Dockerfile

Hey! how are you? I'm sharing this code that allowed me to create and run Facebook's Faiss-CPU 😎 

It's an amazing improvement in time terms vs Sklearn's Kmeans, 10-30 times faster!
FROM continuumio/anaconda3:2019.10

RUN wget --no-check-certificate -O /tmp/mklml.tgz https://github.com/intel/mkl-dnn/releases/download/v0.12/mklml_lnx_2018.0.1.20171227.tgz && \
    tar -zxvf /tmp/mklml.tgz && \
    cp -rf mklml_*/* /usr/local/ && \
    rm -rf mklml_*

RUN conda install faiss-cpu -c pytorch
Like Comment

[5 topics + 3 bonus] Beyond Clustering in Sklearn | Publishing competition

Hey! how are you? I'm publishing this notebook to share some libraries, tips and tricks for clustering problems.

I'm skipping all the explanations that you can find in Sklearn, because the idea is share my current tools in a quick way. Even with this point made, everyone with some interest or experience in clustering is welcome to talk or comment πŸ˜€

These are the 5 topics:
  1. Metrics and Graphics standardization with Scikit Plot
  2. Dimensionality reduction with UMap
  3. Interpretation of clusters with Skope Rules
  4. For times when the dataset keeps growing and your resources stay the same you call Faiss
  5. Demo-ing a webpage (so to speak...you'll see!) with Streamlit directly from Deepnote learned from here

And the 3 bonus are:
  1. Draw data (the cover image origin!)
  2. A Dockerfile to smoothly run FaissΒ 
  3. The github repo (this one is a little bit humble πŸ˜‚)

I'm also entering the Publishing competition, good luck to everyone in it!

Ps: just in case, here it goes the link againΒ https://deepnote.com/@tomas-bonfiglio/Enhance-your-clustering-projects-T176oVvEQZWXnoZZkusECA


Like Comment