Amy Seeley’s Post

Inspired by #3Blue1Brown's (Grant Sanderson) recounting of an interesting way to approximate the irrational number pi, which is four times the sum of the sequence 1-1/3+1/5-1/7+1/9-... (known as Leibniz's formula for pi), I felt I had to see this for myself. This short #Matlab script and figure shows how this sequence indeed tends to converge to pi as the infinite sum is approximated with increasing terms in the sequence.

Our paper, "On the Effect of Quantization on Dynamic Mode Decomposition," has been accepted for presentation at IEEE CDC 2024. Dynamic Mode Decomposition (DMD) is a widely used data-driven algorithm for estimating the Koopman Operator. The paper delves into how the estimation process is influenced by data quantization, uncovering a key theoretical finding that enables the accurate recovery of unquantized estimates from quantized data. This analysis marks the first exploration of the impact of quantization on data-driven algorithms, offering new insights into how communication constraints can affect learning algorithms, particularly in the realm of machine learning. Explore the full preprint at https://lnkd.in/gMrj6q5i. Gratitude to Dipankar Maity for conceptualizing the problem and Sriram Narayanan for the codebase.

A Bird’s-Eye View of Linear Algebra: Orthonormal Matrices Orthonormal matrices: the most elegant matrices in all of linear algebra. Continue reading on Towards Data Science » https://lnkd.in/dqjVDC9D

⭐ Excited to share our latest research paper on manifold data augmentation for tabular/table data titled "TabMDA: Tabular Manifold Data Augmentation for Any Classifier using Transformers with In-context Subsetting". In this paper, we address the challenge of limited tabular data, which often hinders the performance of machine learning models. We introduce TabMDA, a novel method that utilizes a pre-trained in-context model to perform label-invariant transformations and enlarge the training dataset in the encoder's manifold space. Our method is training-free, applicable to any classifier, and improves performance substantially on small datasets. This was a great collaboration with Adrián Bazaga, Nikola Simidjievski, Pietro Lio' and Mateja Jamnik Preprint available: https://lnkd.in/dzdypRjf

Data plots that speak for themselves The latest release of #SigmaPlot 16 is here! The new version introduces advanced tools to make statistical analysis more precise and visually striking. What's new? • Violin Plots: For a deeper analysis of data distributions • Butterfly Plots: For intuitive side-by-side dataset comparisons • Confidence and Prediction Bands: To visualise uncertainty in regression models • Enhanced Error Bars: For better precision visualisation • Excel Multi-Sheet Import Macro: To streamline large dataset imports. From analysing complex biological data to visualising engineering tests and interpreting social science surveys, these new tools can significantly improve your research objectives. Download a free trial today to experience the powerful new features of SigmaPlot 16! https://lnkd.in/dVXUFjAM #statisticalanalysis #violinplots #butterflyplots #confidencebands #predictionbands

I did this video especially for my fellow data science students at Wits :) How to use the matrix mode on the Sharp EL-W506T calculator. https://lnkd.in/dV4MAVEg (the matrix background on the cover page is because I am a nerd 😉 )

Don't one-hot encode your categorical features! If you are using GBDT then both LightGBM and CatBoost get performance gains by using dynamic splitting via Fisher grouping (https://lnkd.in/dNT-Xnug) as long as you previously nominate your columns astype('category'). XGBoost can do this too but is rather timid, and by default has the parameter max_cat_to_onehot=None (so unless explicitly set then XGBoost will use suboptimal one-hot encoding for everything). #datascience #machinelearning #tabulardata

In the world of mathematical modeling and analysis, the ability to efficiently evaluate complex functions over a set of inputs is critical. Unlike Exce­l's drag-and-fill method, Mathcad Prime provides a cle­arer and math-centric way to deal with intricate­ equations and vectors. By defining your function and input vectors separately, you maintain clarity and precision in your mathematical expressions. In this blog, you'll learn the process of defining a complex function and evaluating it in Mathcad Prime: http://ptc.co/n99O50R1s2i

Looking forward to another edition of the two-day course about Linear Mixed Models in R. We are very happy to work together with Joris De Wolf since several years! If you are interested in the next edition, more info on the course: https://shorturl.at/jO148 #technology #training #statistics #linearmixedmodels