Causal Analysis of Corner Kicks
A causal framework for the analysis of Corner Kicks, using the combination of Opta tracking & events data.
I am currently working as a Machine Learning Engineer at KnetMiner integrating of AI and Machine Learning to the KnetMiner platform and the bioinformatics space.
I hold a Masters degree in Computer Science alongside a comprehensive skill set acquired during my time with KnetMiner and previously as a Software Engineer at Hewlett Packard Enterprise. My professional journey has equipped me with valuable skills in automating infrastructure, utilizing RESTful APIs, and deploying in a CI/CD environment.
Additionally, my academic pursuits have focused on applying Machine Learning algorithms to Vision and NLP datasets. My Master's Thesis on training Score-based generative models with Non-Gaussian noise can be found on Indigo-UIC. The research was also selected for poster presentation at the MMLS 2023 conference. You can find the pre-print version of this paper on arxiv .
You can find my football analytics articles on my blog. My data driven pieces on Tottenham Hotspur Women can also be found on the Spurs Women Blog.
If you want to know more about me or my work, please don't hesitate in reaching out via, email: mishraharsh169@gmail.com
Resume Google Scholar Football blogCausal Analysis of Corner Kicks
A causal framework for the analysis of Corner Kicks, using the combination of Opta tracking & events data.
Detection and analysis of corner situations using only broadcast tracking data.
Used standard data pre-processing and cleaning techniques, applied K-means to cluster player locations into zones.
A causal analysis pipeline to identify factors that influenced public sentiment during the COVID-19 pandemic.
We present a comparison between NoTears, NoCurl and NoFear algorithms to find causal graphs and then use Bayesian Networks to find the conditional probabilities.
Training Score-based generative models with Non-Gaussian noise.
We present a novel architecture and show its utility in image generation, point cloud denoising and generative PCA.
End to end Log Analysis Pipeline.
AWS EBS - generate logs, Hadoop Spark & MapReduce - data crunching, Kafka - real time streaming of logs, D3.js - real time visualization.
Interactive visualization of protein sequence data - D3.js
Used SDLC and data cleaning & visualization techniques to come up with ways to visualize amino acid protein data.
Algorithm to convert categorical labels/features to continuous labels.
The algorithm allows the use of kernel methods for node classification and other GNN tasks. Applications also on Event Stream and Pose Estimation data.
End to end tweet extraction and summarization.
Extracts tweets for a given keyword, in a given time period, and produces a concise summary along with the tweet sentiments. Model available on hugging face.
Find similar soccer players using UMAP & GMM.
Applied UMAP to convert soccer events data to smaller dimension, then compared GMM's clusters to actual player positions.
Open source contribution.
Made the original code compatible with any given dataset, fixed versioning issues, and changed the output format required by "Controllable Video Generation with Sparse Trajectories", CVPR'18.
If you want to know more about me or my work, please don't hesitate in reaching out.