Nikhil Thorat

Lilac Co-founder of Lilac AI. Co-creator of TensorFlow.js and Know Your Data. Google Brain, PAIR, Responsible AI.

Results 53 issues of


Nikhil Thorat

Improve UX around schema menu with computed signals / concepts

Currently it's a mess: Let's clean this up a bit more so it's more opinionated.

[UI] Double render bug of embedding index: parent of two fields with an index.

See https://lilacai-lilac.hf.space/datasets#lilac/opus100-en-es-validation

Show concept model metrics in the concept pill.

When searching by a concept it would be great if we could show the concept metrics with the pill, for the given embedding.

Add a Kaggle source.

Doing this will allow us to avoid hosting the_movies_dataset.csv dataset.

Show a message when a dataset is sampled.

We currently don't show a message saying if a dataset has been sampled.

When refining a concept in a dataset view, the skeleton text appears but the old results still are there.

We should hide the current results.

Add source code links from signals.

We should link to the source code for each signal in the signal preview page.

Add a flag to HuggingFace source to load every subset together.

This will let us load popular datasets like blimp without having to run many datasets: https://huggingface.co/datasets/blimp

Enable parallel CSV reading after multi-line reading in duckdb is fixed.

Turning on parallel reading with multi-line CSV rows (e.g. a quoted string on multiple line) messes up the parallel reader. We should file an issue against duckdb with a minimal...

Rename "leafs" to "petals", which are nodes with values.

Currently, schema has "leafs" which returns all the nodes with values. This is technically incorrect, so let's introduce "petal" terminology for nodes with values. Leafs are truly leafs.