Labs¶
Experimental and in-progress notebooks live here. These are works-in-progress that explore new ideas, prototype features, or test approaches that haven't yet graduated into official tutorials. Expect rough edges.
Oval — Opinion Variable Annotator¶
An analytical tool for examining Pol.is-style crowdsourced opinion data. Define variables of interest, then use human-annotated "anchor" comments to approximate those values across all participants — building on top of Polis voting pattern analysis.
React App
Polis Report Processor¶
A streamlit app for taking any polis report URL, running some valency-anndata processing, and exporting results as an h5ad export. Basic form fields allow customizing the data processing.
Streamlit App
Perspective Map Explorer v2¶
Like Google Maps, Polis, and MS Paint had a love-child. Allows exploring h5ad data exports produced by valency-anndata. Paint groups with colors, then click colors to view statements that differentiate groups. Or animate between different dimensional reduction projects, and view any participant data layer you've created.
React App
Semantic Statement Map Creator¶
A streamlit app that uses the Polis 2.0 pipeline for generating explorable semantic maps of conversation statements. All you need is a Polis report URL. Simple translation via Google Translate. Uses DataMapPlot visualization library.
Streamlit App
Exploring Participant Engagement Metrics¶
Visualize participant engagement time and number of unique activity sessions on the perspective map projections. These might be added as future helper methods in valency-anndata.
Python Notebook
Vote Heatmaps and Clusters¶
Visualize Polis vote matrices as heatmaps with participants grouped by various clusterings like KMeans or Leiden. Visually compare how structure of this data compares between detected groups.
Python Notebook
Embedding Reliability via ZADU¶
Evaluate the quality and reliability of dimensionality reduction embeddings (PaCMAP, LocalMAP, UMAP) using the ZADU library. Explore how well each projection preserves the structure of the high-dimensional vote matrix.
Python Notebook
Clearer Maps with Highly Variable Statements¶
Explore how filtering to highly variable statements sharpens participant perspective maps. Using the Chile protest Polis dataset, this notebook compares PaCMAP and LocalMAP embeddings built on all statements versus the top 50 and 100 most dispersed statements — showing how statement selection affects the clarity of group separation.
Python Notebook
Custom Notebook Tools via Anywidget¶
Embed custom interactive react components directly in Colab notebooks using anywidget. A proof-of-concept for building bespoke exploration widgets alongside valency-anndata analysis outputs.
Python Notebook