HASH vs other tools for building simulations

Gena Mal over on the HASH Slack asks:

Hey guys, it’s good to have Hash popping up in the simulation industry as a new tool since the area has traditionally been developing quite slowly. From my observations, the other tools on the market are moving towards simplification of model-building for industry specialists (towards no-code or almost no-code), because pure programmers are often not the best simulation developers. And some of the established players in the industry like Anylogic, Simio, Flexsim, Simul8 put a lot of effort into creating prebuilt functionality for industry-specific projects, like manufacturing, warehousing, etc. To replicate these industrial strength tools’ features would take years. What do you think are the traits that will make Hash stand out?

We replied:

Hey Gena, thanks for popping in! We’re of a shared opinion that domain experts are the best builders of simulation models, and that barriers to simulation should be lower. That’s why we’ve built hIndex as an open community of simulations and their component parts.

On Monday 29th March (8 days from now) we’ll be publicly launching our HASH API (currently in private beta) which allows for the creation of custom user-interfaces atop HASH. We’re working with a number of external users already to craft industry/problem-space specific interfaces for building simulations. If there’s a particular area you’re interested in, we’d love to learn more. Please feel free to DM me.
Re: “industrial-strength” tooling, HASH was borne of a recognition that existing tools, simply put, were not in fact industrial-strength. We wanted to support the ingest of streaming data, the scaling to millions of agents, the distributed compute of simulations, and the ability to programmatically access results — all within a universally schema-mapped, normalized data environment. That’s not something any other tool offers today, and it was the original reason we began creating HASH.

Gena replied:

Hey David, thanks for the reply. You’re right to say that scale is an issue with many other tools. Real models often require running on server or in the cloud. Although purely “distributed simulations” where you slice your model into multiple smaller models and run it in parallel is a bit of a dream of a bright future to me. Can you also expand a little more (or point me to a link) about “programmatically accessing results”? What’s going to be so special here?

Talking about industrial-strength, I mean the ease of model building process. E.g. in Flexsim there’s a lot of visual drag-n-dropping, with blocks heavily geared towards the material handling industry, and the animation works out of the box. Anylogic is more agent-based focused (I bet that’s something you’re looking at intensely), and they also have cool libraries of pre-built behavior for pedestrians, rail, etc.

Of course there’s a lot that you have to do by coding in these tools, but they’re getting easier over time - I bet a lot of this comes from user feedback. From my experience, it’s been the biggest barrier for industry specialists. And Hash seems to be purely js/Python coding. Python is the standard for data science, but still this is coding, not building blocks. Do I assume it right that you expect more of data scientists come into the field?..

We responded:

Hi Gena, re: programmatic access, while a limited subset of other tools provide means for accessing run data from remote environments, they don’t provide:

  • APIs for triggering runs and retrieving the resulting data;
  • the ability to set up schedules for simulations to be run on (linked to changing external datasets); or
  • webhooks which allow HTTP push notifications to be received whenever data is created or updated.

We intend to support all of these cases for integrating HASH models into operationalized business processes.

Re: drag ’n’ dropping by providing our API we hope to see plenty of domain specific user-interfaces emerge. We’ll be building a number of generic ones ourselves for system dynamics and process model building, but there are niche consultancies using HASH in a variety of industries who’ve expressed an interest in building their own plugins, and making them available.

Re: Python support, one of our objectives is to make distributed agent-based modeling something that is within reach of any competent data scientist today. Our focus on engineers first reflects our intention of getting the primitive underlying structures right first before layering atop user interfaces. Too many competing tools have rushed to GUIs and locked themselves into poor path-dependent routes that have hamstrung their ability to generalize down the line, and we do not want to repeat their failings.Feel free to DM me if you have further questions!

We’ll post any further updates here so they’re not walled within the Slack, for others who wish to follow along and are curious about alternatives to AnyLogic, Simio, Simul8, etc.