Why this model is called non deterministic?

Could someone explain me what exactly is non deterministic.

There is another photo of deterministic model

You posted this in the General Discussions category. I think it belongs to the Probability course, right? Please change it to the correct course category. Read this on how.

It is not linked with any course. i am asking it generally.

If you think it should be added to different group, please help me finding its name.


For general queries, this is the correct place.

Regarding your question, I don’t know about it. But I hope someone else will reply to you soon.

This seems a little unfair: you’re taking some other course that has nothing to do with DLAI and then you’re asking us to explain the lectures of that other course to you without giving us any real context for that they are talking about …

But I think it’s actually straightforward, if you just look at the difference between the two models. Deterministic means that the same thing happens with the same inputs every time. Non-deterministic means that I can feed the model the same input and not necessarily get the same output every time.

Look at the difference in the two models that are built there: in the deterministic case, they just feed to input to a sigmoid function, so you get whatever the output of sigmoid is for that input, right? So it will be the same every time with the same input.

In the non-deterministic case, they first feed the input to sigmoid and then they take that output and feed it to a Bernoulli Probability Distribution function. Now of course it depends on how that TF function works, but my guess is that you don’t necessarily get the same answer if you feed it the same input multiple times. Please read the documentation for the function and I hope it will discuss that point. Perhaps it is possible to get the same output by setting the random seeds the same way every time, but then what’s the point of using a PDF function in the first place. Presumably the whole point is that they are showing you how to introduce non-determinism into the behavior, although you have given us no context for understanding where they are going with this idea.

1 Like

In addition to Pauls‘s excellent reply: just have a look at the blue scatter plots, @tbhaxor:

  • In one case you see that there are just two possible outcomes (true or false; encoded with 1 or 0). These are the data sampled from the Bernoulli probability distribution, which is why the model is probabilistic.

  • in the other case you get just the deterministic outcome of the model (due to its activation layer choice) of the sigmoid function which you also plotted with its formula in this same viz and can see they match quite perfectly.

Best regards