# Rust ML WG Meeting 00023 ## Meeting Info Date: 20210414 Start time: 1400ET Zoom: https://ucsb.zoom.us/j/6601852842 ## Agenda - DBSCAN chapter has been added to rust-ml book - Starting work on ICA next - Lorenz and Manuel added oxide-enzyme to the rust-ml Github repository. Highly WIP, but shows how to generate gradients with Enzyme. - no type-tree generation for now, but got simple `f(x) = (x^2+1) + 2x` working (with nested functions) - provide starting point for Manuel's thesis - Give a small showcase? - Linfa - incremental KMeans implemented (@YuhanLiin) - ~~almost ready for merging~~ merged: `linfa-preprocessing` by Ivano (@Sauro98) - Platt scaling and multi class composing model needs review - bump `linfa = 0.4`? ## Participants - Chris - Manuel - Ricky - ivano - William - yuhan - Lorenz ## Minutes - Meta - Been a few weeks - woo good number of people for this meeting! - DBSCAN chapter has been added to rust-ml book - :tada: - independent component analysis next - should go quicker - Lorenz and Manuel added oxide-enzyme to the rust-ml group - tried to integrate enzyme into Rust ecosystem - basic prototype with type signature function for a heard start for manuel's thesis - meets with supervisor tomorrow to announce thesis :clap: - Screen sharing - and poking through the code - any pre or post processing optimization? - no not yet - you want all of the pre-optimizations on - except vectorization and unrolling - might want to enable post-processing optimizations on - Linfa - yuhan implemented an incremental k-means - ivano merged `linfa-preprocessing` :tada: - lorenz implmented platt scaling and multi-class composing model - needs review though - some examples are bad and could use some help digging into where it fails - probably publish next linfa once platt scaling is merged - blocked on the ndarray version with sparse matrix library in the trait system - chrism writing licnensing artice for _The Gradient_ - machine learning, ethics, and open source licensing - teneatively accepted for publication (parts 1 & 2) - released over the next 2 or 3 weeks :tada: - bio should say rust-ml WG :D so hopefully some exposure there! - question on how does the new testing benchmarks for multi-thread/processing applications - not totally sure about this with cache misses etc, but should always take results with a grain-of-salt ## Action Items - chrism review the platt scaling MR - https://github.com/rust-ml/linfa/pull/112 -