The paper includes a comparison to TabPFN v1 (among others), noting the lack of categorical & missing values handling which v2 now seems to have. Would be curious to see an updated comparison.
TabPFN is better on numerical data since v1 (see figure 6 in the CARTE paper). CARTE's main strength in on text features, which are now also supported for TabPFN v2 API version (https://github.com/PriorLabs/tabpfn-client). We compared this to CARTE and found our model to be generally quite better, and much faster. CARTE multi-table approach is also very interesting, and we want to tackle this setting in the future.
Largest Contentful Paint [1], one of the main metric to measure perceived load speed, tends to penalize low-quality image placeholder. This article digs into how, what can be done to remain on the right side of LCP when your LCP is an image, and why you don't want to just upscale.
We make a software that we then sell. It's a data science platform that customers install in their internal IT, that allow various profiles from data scientists to business analysts to bring their skills to their data projects with both user-friendly interfaces and code.
Interesting problems to solve and plenty of learning opportunities.
https://soda-inria.github.io/carte/ https://arxiv.org/pdf/2402.16785
The paper includes a comparison to TabPFN v1 (among others), noting the lack of categorical & missing values handling which v2 now seems to have. Would be curious to see an updated comparison.