@C.E. and other, is it worth reporting that HTTPRequestData[] can contain None values which makes it impossible to automatically export as JSON? A format one could expect it to be compatible with. If HTTPRequestData[] could contain Null for those cases, then it exports well :-/

@Kuba I don't know, I understand your point but I think None is the correct label from a Mathematica point of view...

@C.E. Null exists too, and it is documented as an "absence of an expression or a result"

Has anyone else watched the virtual seminar on Neural Networks one week ago?

ave you received the notebooks in email?

@Kuba Which is wrong on so many levels :) But for the general user it is maybe a reasonable explanation.

@halirutan I like to abuse documentation wording when I write bug reports :)

Any sources about that change anyone?

@Kuba But there are many cases where Null does not work as None. Although both specify, in some sense, the absence of something, they are not interchangeable. Note that HTTPRequestData has to be compatible with HTTPRequest, which I think is one of those functions where None and Null are not interchangeable. Probably justifiably so since Null and None are not, in general, interchangeable.

@C.E. I don't insist on Null, it is just ridiculous that you need to take special care to convert request data to JSON.

@Kuba If I look at the 11.1 and 10.4 documentation side by side, the details options seems to have been updated with much of the new behavior I think. Does that help at all?

@JimBaldwin I probably ask this anywhere, but maybe you can give me a hint like last time. I have experimental data where measurements where ranked. So I have an ordinal variable but for simplicity let's assume the following example: I have two unfair dices and I don't know which one is better. I, as observer 1, throw each dice 50 times and I get the counts for each number and each dice.

I know how to test if the dices are statistically different, but what I'm interested in is which dice is better. What about order statistics do I need to know to find out which one it is?

My second question is: Lets say I have an observer II that also throws each dice 50x. Can I create a combined outcome for dice one and two, taking into account that observer II might throw the dice in a different way?

I guess a more intuitive example would be two classes of 50x pupils each and two teachers who give the pupils marks.

@halirutan Because of the dice analogy, I'm assuming that you have a relatively small number of outcomes (for the dice 1 through 6 or A, B, C, D, E, and F with A<B<C<D<E<F) with true but unknown proportions p1, p2, p3, p4, p5, and p6. The first thing to do is to define the metric that characterizes what you call "better". That might be 1*p1+2*p2+3*p3+4*p4+5*p5+6*p6 (i.e., the mean rank). Then I can suggest how you might use the data to see if one die is better than the other die.

These 2 "Geo" bugs (1, 2 seem to be fixed since v11.1.0 (tested with OSX). May someone confirm with Windows OS and update the post ?

For the second question likely some "generalized linear model" or "generalized linear mixed model" would be appropriate. (Mathematica does not directly perform the latter.) The recommended analysis (or at least characterizing the scope of the inferences) depends on if the data available consists of 100 students randomly assigned to two teachers or if there was some other selection criteria used to assign students to teachers.

(In other words my standard statistician's response: "It depends.")

@JimBaldwin Yes, it is a small number of categories 0, 1, 2, 3 where 0 means unstructured and 3 means highly structured. The numbers are given by the observer and even if there are true proportions, we don't know them. We only have the ranking that each observer gives a measurement.

The question about better is what drives me crazy.

Assume we have 6 samples from a probe and here we have 3 with a value of "0" and 3 with a value of "3". Then let's assume a second probe with 3 values of "1" and 3 of "2". They are clearly different but it seems to be hard to define which is better.

(my example in the last paragraph can be translated having two dice and throw each 6 times. One gives three times a 1 and three times a 6. The other gives three times a 3 and three times a 4.)

@JimBaldwin So when I understand you correctly, I have to define the measure of goodness myself?

@halirutan There is no information in a set of numbers about what one should do with the numbers. If there was, then we statistician's wouldn't need you physics/engineering types. So, yes, my recommendation is to think as to what a good characterization should be. Or you could take the default category mean. But, of course, the "distance" between category 0 and 1 is not necessarily the same as the distance between 1 and 2.

It's OK with your subject matter knowledge to assign distances. (The catch is getting your colleagues to agree with you.)

@JimBaldwin OK, that sounds reasonable. Having such a mean measure would instantly solve to problem how to combine the evaluation of two observers for one probe.

I'm giving this some more thought and maybe I hit over to statistics.se and ask a question with a more real example. This kind of "we have rated something" problems arise often in medical experiments, where scientists can only measure the quality of e.g. cell growth (or growth quality) subjectively.

In any case, thanks for your input.

@halirutan Accepted measures evolve over time. For instance, maybe categories 0 and 1 aren't so important and the proportion of 3's and 4's is what wakes someone up. Further one could have a "2 dimensional" assessment rather than a single number. Maybe the mean and the proportion of 4's. The point is that there is no single right measure.

Newly proposed measures need to gain acceptance and are best validated by desired consequences coming true.

@user6014 No it does not considering the volume of documentation. I didn't know PolyhedronData went under backward incompatible changes untill my code broke. Should I compare whole documentation of updated symbols?

why is it that, e.g., QuantityUnit@Quantity[5,"MilliKelvins"] evaluates to MilliKelvins but milliKelvinize[t_Real]:=Quantity[t,"Millikelvins"]; QuantityUnit@milliKelvinize[5] remains unevaluated?

@Kuba Sorry, didn't mean for it to be a silly suggestion. Clicking on the "Updated - Show changes" icon at the top relatively clearly (assuming it is complete) shows & hilights that there were a lot of changes. It would be nice if there were a more direct listing, and explanation, of the changes. I understand your frustration :/

@user6014 sure, no worries. I am aware of "show changes", the problem is that changes/updates are usually backward compatible, and PolyhedronData wasn't just some experimental function, it was around for quite some time.

@JulianWolf If a function call stays unevaluated, it usually means two things in Mathematica:

The first case usually happens with built-in functions and if you ask such a question, it is definitely not your problem, since you really have to implement this behaviour. This leaves you with the second case for your problem. Therefore, can you think why your call remains unevaluated while this here works:

QuantityUnit@milliKelvinize[5.0]

If you understand this, then you have your solution.

After that, you might want to have a look at NumericQ or NumberQ and decide what you rather should use.