4 ms·

Less Scatterbrained Scatterplots

28 points by asah 6y ago

setr 6y ago
So this is just... plotting with zoom levels and aggregations? It seems to me the only novelty is to attach zoom to the scrollwheel, which I suppose is less common in BI visualizations (often a button instead)
- jamessb 6y ago
  The novelty is not the scroll-to-zoom interaction - it's the creation of a declarative grammar for several visualization types (see Figure 2 for examples), and a system that implements this in a way that is scalable to datasets containing billions of observations. The contributions claimed in the paper are [1]: > To summarize, we make the following contributions: > •An integrated system called Kyrix-S for declarative authoring and rendering of SSVs at scale. > •A concise and expressive declarative grammar for describing SSVs (Section 4). > •A framework for offline database indexing and online serving that enables interactive browsing of large SSVs (Sections 5 and 6). [1]: https://arxiv.org/pdf/2007.15904.pdf https://arxiv.org/pdf/2007.15904.pdf
  - dlivingston 6y ago
    Very cool. This could give Plot.ly a run for its money in the data science community. Also regarding Figure 2, and at the risk of sounding cliché: my inner graphic designer is weeping at the use of Comic Sans as their choice of code font.
- tracyhenry 6y ago
  First author here. In addition to a declarative grammar, a big thing is the ability to scale past browser memory limits. The system works with a distributed Postgres (Citus) to perform parallel hierarchical clustering to get the aggregation computed, and uses Postgres spatial indexes to fetch data on demand. We can scale to billions of records (see reddit comment demo in the news article). Also we are not limited to common representations such as circles. In fact you can create your own in D3. Here's a live demo of 1.88 million historical wildfires: http://wildfire2.kyrixdemo.live/ http://wildfire2.kyrixdemo.live/
niming2 6y ago
Looks like a major upgrade to leaflet.markcluster (https://github.com/Leaflet/Leaflet.markercluster https://github.com/Leaflet/Leaflet.markercluster), which supports 50k points at most.
jeremytech 6y ago
This is certainly a promising endeavor.
eternal_answer 6y ago
There seems to be a missing gif (the improvement for NTY article)? I wonder what it looks like
vote4myboi 6y ago
Very cool initiative
kkytle 6y ago
The declarative grammar is cool (flexibility) but the reason we care is because it can SCALE. No more counting points. Forget browser memory limits. BILLIONS of records.
observercs 6y ago
hey nice work. It seems like the hierarchy of observations across zoom levels is prematerialized in a database. How would that impact the ability of the system to handle stream data?"

28 points by asah 6y ago

setr 6y ago

So this is just... plotting with zoom levels and aggregations? It seems to me the only novelty is to attach zoom to the scrollwheel, which I suppose is less common in BI visualizations (often a button instead)

jamessb 6y ago
The novelty is not the scroll-to-zoom interaction - it's the creation of a declarative grammar for several visualization types (see Figure 2 for examples), and a system that implements this in a way that is scalable to datasets containing billions of observations. The contributions claimed in the paper are [1]: > To summarize, we make the following contributions: > •An integrated system called Kyrix-S for declarative authoring and rendering of SSVs at scale. > •A concise and expressive declarative grammar for describing SSVs (Section 4). > •A framework for offline database indexing and online serving that enables interactive browsing of large SSVs (Sections 5 and 6). [1]: https://arxiv.org/pdf/2007.15904.pdf https://arxiv.org/pdf/2007.15904.pdf
- dlivingston 6y ago
  Very cool. This could give Plot.ly a run for its money in the data science community. Also regarding Figure 2, and at the risk of sounding cliché: my inner graphic designer is weeping at the use of Comic Sans as their choice of code font.
tracyhenry 6y ago
First author here. In addition to a declarative grammar, a big thing is the ability to scale past browser memory limits. The system works with a distributed Postgres (Citus) to perform parallel hierarchical clustering to get the aggregation computed, and uses Postgres spatial indexes to fetch data on demand. We can scale to billions of records (see reddit comment demo in the news article). Also we are not limited to common representations such as circles. In fact you can create your own in D3. Here's a live demo of 1.88 million historical wildfires: http://wildfire2.kyrixdemo.live/ http://wildfire2.kyrixdemo.live/

niming2 6y ago

Looks like a major upgrade to leaflet.markcluster (https://github.com/Leaflet/Leaflet.markercluster https://github.com/Leaflet/Leaflet.markercluster), which supports 50k points at most.

jeremytech 6y ago

This is certainly a promising endeavor.

eternal_answer 6y ago

There seems to be a missing gif (the improvement for NTY article)? I wonder what it looks like

vote4myboi 6y ago

Very cool initiative

kkytle 6y ago

The declarative grammar is cool (flexibility) but the reason we care is because it can SCALE. No more counting points. Forget browser memory limits. BILLIONS of records.

observercs 6y ago

hey nice work. It seems like the hierarchy of observations across zoom levels is prematerialized in a database. How would that impact the ability of the system to handle stream data?"