There are already many wonderful and successful companies that offer, as we do, Bayesian Network (bnet) software and services (for instance, www.bayesia.com, www.bayesfusion.com, www.bayesserver.com, www.hugin.com, www.agenarisk.com and www.norsys.com). So what makes us different and better?
Most of our competitors follow a 20 year old business model in which one sells expensive subscriptions for the use of secret, proprietary software. So if you write a bnet with their software, it will be tied (“locked-in”) to varying degrees to their proprietary formats, plus you are at their mercy for perpetuity as far as pricing is concerned. Instead of that, at Artiste we use only free, open source software. For example, we are experts at :
- azua (Microsoft)
- books that contain excellent python code
- causalimpact (Google)
- CausalLib (IBM)
- CausalML (Uber)
- DoWhy (Microsoft)
- JudeasRx (by ar-tiste.xyz, BY US)
- PyLift (Wayfair)
- PyTorch by Facebook and its probability adjunct Pyro by Uber
- Tensorflow (especially its probability module) (Google)
and many other bnet related software libraries. If you write your bnet using these fantastic open source software libraries, there is much less lock-in. Open source software has the additional benefit that it is much more flexible. You can modify the code to suit your needs if it is open. You cannot modify easily that which you cannot see. Another benefit of open source software is that it is bleeding edge and is part of a vast software ecosystem. Do you really want to limit yourself by using only one piece of software written by a small company that employs a mere handful of good programmers, or do you want to use the latest and greatest software coming out of hundreds of companies and universities from around the world, including Google’s Tensorflow?
Another advantage of Artiste over our competitors is that, unlike them, we use Jupyter notebooks to the fullest extent possible.
Many of our competitors will provide you with the binaries for a GUI-rich app (GUI=Graphical User Interface) that runs on your PC. This is the old, traditional model of distributing software. To be fair, there is much good to be said about GUI-rich apps. GUI’s excel at reducing the possibility of user errors, increasing the ease of use, and reducing the amount of understanding of the code that is demanded from the user in order for him or her to use the code correctly.
We think Jupyter notebooks (nbs) are much better than GUI apps. nbs allow the user to save every step of a calculation, including numerical results and graphs, and to share it with others, something that GUI apps don’t normally do. With “widgets”, nbs can also have a GUI.
Notebooks are not a panacea. They can look impenetrable sometimes. The trick to good nb writing is to put as much code as possible, especially the internal code for complicated functions, outside of the nb, and to just call, not define, those complicated functions in the nb.
In our opinion, nbs are the best way ever invented for documenting your work in Data Science. Documenting your work is very valuable to a business. This is especially true if you have to go through several iterations of your algorithm to get it to work, as almost always is the case. Or if you or someone else has to remember what you did a month ago.
Another advantage of nbs over GUI apps is that nbs are available for most computer languages (Python, R, Octave/Matlab, etc.) so no programmer or software package is left out due to language barriers.
Another advantage of Artiste over our competitors is that, unlike them, we are very quantum savvy. We offer software and services for both classical and quantum bnets. We are highly capable of writing quantum computing software for your company, if that is what you want.