How people browsing on NASA website on 90’s

What are we talking about ? When websites were still in their infancy, we were able to reconstruct how early Internet users navigated one of the first versions of NASA’s website intended for the general public. It is from the Apache access log of July and August dating from 1995, that it was possible to … [Read more…]

Application of MCTS within the Connect4 Game

  Popularized by Deepmind, a subsidiary of Google specializing in artificial intelligence. The MCTS, Monte Carlos Tree Search, is the algorithm used in combination with a convolutional network, which managed to beat world number 1 Lee Segooe in the game of Go. Before being, again overtaken by the AlphaGo Zero another algorithm designed by Deepmind. … [Read more…]

Visualization of scikit-learn Decision Trees with d3.js

Machine-learning algorithms are often criticized for the opacity that surrounds them. It is difficult to understand how predictions are evaluated. The way in which supervised algorithms work, no matter how complex, is known. But it remains difficult to understand how one observation is predicted in one class rather than another. The inaccessibility of the characteristics … [Read more…]

Computation of the Moore-Penrose inverse with CUDA / C ++

  The Moore-Penrose inverse matrix or pseudo-inverse matrix is defined by: It especially used in the normal equation to determine the coefficients of a linear regression. In this context, it is possible to calculate the results in different ways by using an optimizer algorithm that approaches by iteration  the optimal result or by using the … [Read more…]

Segmentation : Overview of the state of the art

  Segmentation is a widespread analysis, bringing particular advantages to companies. It can be used in very different fields and sectors of activity such as health, industry, human sciences. Or in the top management too, marketing or finance departments for instance. In this article, we will focus on a generic methodology that adapts to the … [Read more…]

Deep-learning: Rooftop type detection with Keras and TensorFlow

[version française] With the emergence of Deep Learning, Neural Networks predictive capacity looks extremely promising. This can be seen in the amazing number of theoretical works and applications in the field in the last years,  reaching precisions never observed before. The objective of the article is twofold, first we present  a new dataset of rooftop … [Read more…]

Bubble chart with D3.js

    Version française “Bubble”-plot is a different way to present your data, this kind of plotting broaden the possibilities of representation beyond the classics histograms and pie charts. The program is based in a open-source script which has been modified to be able to add other features, for example values in percentage ,the feature’s … [Read more…]