Hi. I am Mauro

A pythonic data scientist, with an MBA
and a passion for interactive visualizations in D3.js

Projects and investigations

In the next sections you will find a sample of my works

  • Developed during my bootcamp at Metis:

    Machine learning and deep learning projects with some visiualizations in D3.js

  • Developed during my Master in Big Data and Social Mining:

    A project that was published in "Nova" (Il Sole 24 Ore)

  • Developed for ISTAT (Italian National Institute of Statistic)

    A research presented to "SIS2017 Conference Statistics and Data Science"

  • Developed during my spare time:

    Passion projects developed just for the sake of learning

  • Investigations:

    -discrimination threat in machine learning
    -difference between RCNN, Fast RCNN and Faster RCNN

Personality on Twitter

Developed during the Master in Big Data at Univ. of Pisa

Description: Emulation, for the Italian twitter community, of what Cambridge Analytica did in the US. Determined the main traits of personality of 15 million crawled Twitter accounts by applying the Big 5 thoery to their 1 billion tweets. The resuts have been published on "Nova" (Sole 24 ore)


Preparation time:

3 months


Delivery format:

Presentation to professors' commitee


Delivery time:

15 minutes


Applied skills:

  • API crawling
  • Natural language processing
  • Clustering
  • Word2vec
  • D3.js

iPhone

Click on the image to access the presentation
Use the arrows on the bottom right of the page to navigate

Real time facial emotion classifier

Passion project to practise skills in deep learning

Description: Training of a convolutional neural network for real time emotion classification of each face detected, through OpenCV, in a webcam video.


Preparation time:

3 weeks


Delivery format:

na


Delivery time:

na


Applied skills:

  • Convolutional Neural Networks in Keras
  • OpenCV
  • Image processing

iPhone

Click on the image to access the presentation
on Google Drive

Get your project funded in Kickstarter

Developed during the bootcamp in DS at Metis

Description: Trained a 0.85 accuracy classification model to predict, before launching, if a project will be funded in Kickstarter. Best model chosen from 12 different models and an ensemble one built on both numerical and text features


Preparation time:

2 weeks


Delivery format:

class presentation


Delivery time:

5 minutes


Applied skills:

  • 12 fine-tuned classification models
  • Ensembling
  • Natural language processing
  • Principal Component Analysis (PCA)
  • D3.js

iPhone

Click on the image to access the presentation
Use the arrows on the bottom right of the page to navigate

As Shakespeare would say

Developed during the bootcamp in DS at Metis

Description: A seq2seq LSTM Neural Network has been trained to translate current English text to Shakespearean vocabularly and style. A "home-made" text augumentation technique has been designed and applied to extend the limited parallel corpora available after cleansing, thus improving the translation quality


Preparation time:

2 weeks


Delivery format:

class presentation


Delivery time:

4 minutes


Applied skills:

  • Seq2seq LSTM Neural Network in Keras
  • Bidirectional Neural Network in Keras
  • Scraping
  • Natural language processing

iPhone

Click on the image to access the presentation
on Google Drive

Let's (Ted's) talk

Developed during the bootcamp in DS at Metis

Description: Dynamic and interactive web tool to explore 2700 Ted talks from 2004 to 2017. Non negative matrix factorization (NMF) has been used to extract the main topic of each speech. Each topic has then been analyzed by extracting its most significant gorup of words (word2vect) and by evaluating both the evolution of its popularity over time and the typycal style used to deliver it, meant as the typical evolution of sentiment during the group of speeches of each topic


Preparation time:

2 weeks


Delivery format:

class presentation


Delivery time:

5 minutes


Applied skills:

  • Natural language processing
  • Non negative Matrix Factorization (NMF)
  • Sentiment analysis
  • Clustering
  • D3.js

iPhone

Click on the image to access the presentation
Use the arrows on the bottom right of the page to navigate

Target FPR on MINST classes

Techincal assignment in deep learning

Description: Trained a convolutional network and designed a more general decision process to meet a strict FPR threshold on each class of MINST dataset


Preparation time:

1 week


Delivery format:

Skype presentation and code review with a technical committee


Delivery time:

1 hour


Applied skills:

  • Convolutional Neural Network in Keras
  • Statistical approach for solid results
  • Image processing

iPhone

Click on the image to access the presentation
on Google Drive

Real time object detection

Developed as a technical assignment

Description: Research on the evolution of RCCN toward Fast RCNN and Faster RCNN


Preparation time:

1 week


Delivery format:

Presentation to a technical committee


Delivery time:

30 minutes


iPhone

Click on the image to access the presentation
Use the arrows on the bottom right of the page to navigate

Can machine learning discriminate?

Developed during the bootcamp in DS at Metis

Description: First of the 2 investigations durning my bootcamp in data science, dealing with real life examples on how biased machine learning can lead to discrimination


Preparation time:

1 week


Delivery format:

class presentation


Delivery time:

10 minutes


iPhone

Click on the image to access the presentation
on Google Drive

Pagerank: network centrality by Google

Developed during the bootcamp in DS at Metis

Description: Second and last investigation developed at Metis. It describes the measure of network centrality used by Google's websearch engine. Applicable in any network, it is an interesting tool for any data scientist


Preparation time:

1 week


Delivery format:

class presentation


Delivery time:

10 minutes


iPhone

Click on the image to access the presentation
on Google Drive

Thanks for visiting.

May we always have a dream to chase!

iPhone