Techno Blender
Digitally Yours.
Browsing Tag

Matrices

Time Series Data and Markov Transition Matrices

Conceptual overview and practical applicationsImage by Oto Godfrey and Justin Morton from Wikimedia Commons: Free to use under CC-BY-SA-4.0 licenseIn this article, we will look at how reframing time series data using Markov transition matrices can yield interesting descriptive insights as well as elegant approaches for forecasting, backcasting, and the analysis of convergence. Going backwards and forwards in time — just like Doc’s retro-fitted DeLorean time machine in the sci-fi classic Back to the Future.Note: All images…

A Few Properties of Random Correlation Matrices and Their Nearest Correlation Matrices | by Rohan Kotwani | Jan, 2023

Generating Insights from the Solutions of a Particle Swarm Algorithm for Finding the Nearest Correlation MatrixBy AuthorThe purpose of this article is to evaluate the performance of two algorithms for finding the nearest positive semi-definite (PSD) correlation matrix. The first algorithm is a classical method, while the second is a brute force approach. This is relevant for data with high dimensions or missing values, as these types of data can often produce non-PSD matrices. The nearest PSD correlation matrix is useful…

How DeepMind discovered new ways of multiplying matrices using AI | by Federico Peccia | Nov, 2022

Photo by Gayatri Malhotra on UnsplashLast month, DeepMind published a paper where they presented AlphaTensor, an AI algorithm able to find faster ways of doing one of the most common algebra operations: matrix multiplication. But of course, there is a valid question you may be asking yourself: what does this have to do with me? Well, in this article I will explain the contents of the paper and if you stick with me till the end, you will find some examples of why this is such a big breakthrough, and how it will affect your…

Using Sparse Matrices in XGBoost. An alternative way for dealing with… | by Gordon Davis | Oct, 2022

An alternative way for dealing with high cardinalityEmpty space is an opportunity. Photo by Nasa on UnsplashYou don’t have to be involved in data science long before hearing about the algorithm XGBoost, including all the Kaggle competitions it has been used in, with great success. There is also no shortage of great tutorials online (including on Towards Data Science) on how to get started using this algorithm. However, there is an amazing feature of XGBoost that is often overlooked and is sadly missing from most…