Labelling support!

Finally in the clusterfun library: labelling!

This release enables multi-class labelling.

The label tab

• Keeps track of what you have labelled, easy for e.g. a residual error analysis
• Allows you to download specific labels of your selection or all labels of your plot
• Allows you to create new grid views with labelled media
• Allows you to (de)select a label for all media currently visible

Labels are automatically saved, so when you load the same plot again, it'll show the labels again.

Labelling can be done with keyboard shortcuts - just hover over a data point and press the number of the label you're interested in:

v0.4.2

Add audio support and display parameter.

Currently, we determine the media type naively by checking if the file extension is included in:

[
    "mp3",
    "wav",
    "aac",
    "ogg",
    "flac",
    "wma",
    "m4a",
    "aiff",
    "midi",
    "ape",
    "wavpack",
    "alac",
    "ac3",
    "opus",
  ]

Audio is autoplayed on hover.
When displaying audio, the grid view doesn't have an image, but I can imagine it will be useful to display something there.
That's why there's a new parameter display added to all plot types that allows you to display the value of one or more columns in the grid view, as such:

This can be created like this:

import pandas as pd
import clusterfun as clt

df = pd.read_parquet("https://raw.githubusercontent.com/gietema/clusterfun-data/main/libri_speech_test_clean.parquet")
print(clt.grid(
    df,
    title="LibriSpeech test-clean",
    media="filepath",
    show=False,
    display="translation",
))

A less naive way of checking for audio support can be added later if this is not sufficient.

Grid selection download button

The grid now has a download button that allows you to download the selected media as a csv.

Continuous colour scale

Add `is_categorical=False` to a bar chart, histogram or scatter along with a `color=...` to get a continuous colour scale.

Storage client

This release allows you to create different storage clients, so you can more easily create a version of clusterfun that loads data from different sources.
This should open up the way to add GCP support soon as well.

Thanks to @hyenal for adding this!

v0.2.3

Mostly bug fixes in this release:

• Python 3.12 support, dropped 3.8 support

• Fix bug that made it impossible to create plots with two classes

• Grid page does not reset to the first page when enlarging (clicking) an image on a second page

• Slightly more robust filtering

Next up is labelling and saving/downloading selections.

Refactor and grid features release

• Stats feature for grid view, allowing you to quickly look at plots related to the selection of media.
• The grid view is now filterable.
• Refactor of frontend with tailwind and jotai.
• Library now works from within Jupyter notebooks