Meet Narendra Doshi

icon indicating a website link https://www.linkedin.com/in/meetdoshi90/

icon company IIT Bombay icon location Mumbai icon location I like to mess with my code until it stops working

Results 3 issues of Meet Narendra Doshi

Added Heirarchical prior.

3 comment

Added path smoothing to prevent overfitting for deeper trees whilst improving accuracy. Added mse loss for regression on weighted layers. References: 1. [Path smoothing](https://ojs.uma.ac.id/index.php/jite/article/view/3856) 2. [IHT Gradient descent](https://arxiv.org/pdf/1910.13389.pdf)

Fix for #180

This PR proposes a fix for #180 by replacing the existing torch.quantile implementation with one based on torch.kthvalue. Because torch.quantile limits tensors to 16 million elements, it causes problems when...

Error while indexing large datasets like webis-touche using Fast Plaid

10 comment

Error message on torch quantile with FastPLAID I was trying to evaluate web-touche using the evaluation code provided by pylate (1.3.4). I am using FastPLAID index with torch 2.8.0. Any...

bug