3 ms·
I think WASM and WebGPU is a good idea to run "What If" scenarios in browser
In the last years I needed to run quick "what if" tests on real data, multiple times. Some friends (data analysts) told me to do it in spreadsheets, but melted as soon as I pushed more than a few hundred thousand rows. Traditional BI tools... I tried then Tableau but was overkilling
Consider that I'm building a web-based data analysis tool so I tried pushing the whole thing into the browser. At first I thought it would choke, but with WASM for the heavy math and WebGPU for projections and charting, it actually worked. The browser can chew through millions of rows if you stream chunks right, and UMAP/clustering runs smooth enough inside Chrome to be usable.
The hard part wasn't performance, it was UX. People don't want to wire up nodes or think about pipelines — they just want to tweak a number and watch the graphs shift. That forced me to rethink the interface into something closer to a scenario builder than a workflow editor.
The result is Datastripes: https://datastripes.com
Runs entirely client-side, no backend. I'm still figuring out limits, especially memory management and GPU stability across devices. Anyone here has tried pushing WASM+WebGPU this far? Do you found better ways to handle large "what if" workloads in the browser?
- ycombinatrix 1y agoi have no idea what you mean by "what if workload"
- vinserello 1y agoAnd if the analysis could be extremely computationally intensive, consider an LSTM approach. Datastripes effectively addresses this problem by using WASM to build, train, and run the model in a near-native environment, but heavier models or larger datasets could introduce significantly more challenges.
- sargstuff 1y agoWhat type of data "tests"? aka modeling, simulation, statistical regression, etc. Wide variety of "what if" data environments/systems which have web accessible interfaces to backends (that are not just database systems). commerical mathlab, sass open source[0]: scilab; sagemath; numpy; gnu octave; julia; juniper [0] : additional link references beyond "5 open source alternatives to MATHLAB" : https://opensource.com/alternatives/matlab https://opensource.com/alternatives/matlab
- vinserello 1y agoGood question. Let's start saying that Datastripes isn't trying to replace Matlab/Octave/etc. It's closer to a more lightweight scenario builder that runs entirely in the browser. So instead of coding a model, you drop data in, build a flow (filters, regressions, clustering, projections), then tweak inputs and watch the charts update live. We started with the idea of no backend, no install, runs in WASM/WebGPU. Obv, this strategy cannot be a one-size-fits-all, but it covers a lot of use cases in business and educational domains. It's less about heavy numerical libraries, more about quick "what if" exploration for non-technical users or teams that don't want to set up an environment. Think smaller/faster, but also easier to share where data stays local.
- sargstuff 1y agoAso potential for some biomedical research use cases too! Bit easier setup / use than customized spreadsheet with query links to database.