What a bit of FE-programming can do for you:

@Kuba is there a way to make DockedCells accept options from a stylesheet?

It seems as if it ignores CellFrame, CellFrameMargins, etc.

(nevermind it was an inheritance issue)

@Kuba any idea why this happens:

Where like the cells bracket goes bubbled and you can't delete them en masse

any other site-license users get "The activation key ... is not eligible for renewal" when trying to install v12?

despite the key being listed as "Active" for v12 on the Wolfram user portal

I wrote our site license administrator, we'll see what she says

anyone else get a NetGraph error when trying to evaluate TextCases?

Ah, blowing away the cache fixed it

Do you think this question is ill-posed?(mathematica.stackexchange.com/questions/195544/…)

@CarlLange looks ok here

@apt45 Looks good to me, but I know absolutely nothing about the area so I can't help you, sorry!

@kirkus Yeah, I deleted ~/Library/Caches/Wolfram and it started working again 🤷‍♂️

@CarlLange thanks!

@CarlLange hah, good!

@CarlLange didn't see that msg, was distracted by the spinning legs.

I might file a bug on that, since that's at least two of us who've had weird behavior that was fixed by clearing it out. Maybe something they should consider during installation or first-time startup.

I hate when they make things take more screen real estate without making them more readable: E.g. ?**Periodic*` in V12... /:

I complain about it sometimes but there's so much to love about the wolfram language. I created this cool gif using today's actual sun positions and the actual height infomation for ireland in literally six minutes start-to-finish. I've been programming in python for over a decade and I'm pretty sure I couldn't do this in a day. wolframcloud.com/objects/…

@b3m2a1 about brackets, it happened for me recently but I didn't manage to track it down.

I wish Region etc worked with GeoPositions. So much having to do Region[points/.GeoPosition[x_]->x]

eg mathematica.stackexchange.com/questions/195591/…

@Kuba yeah maybe I need to ask support or JF about it

I'll post this here too. Maybe someone has some good input.

@LukasLang Several of the function names in your package now conflict with v12.0 built-in

@Szabolcs Thanks for the reminder - I think the latest commit (no release yet) should already contain an "emergency" solution for this. There are also a few other compatibility issues which I'm working on, I'll upload a release once the most obvious ones are resolved.

@CarlLange One note about gifs, the default is to export them so they only play once. To get them to repeat you have to export with some kind of Animations-> Infinity rule in the Export command.

But that is a really cool gif.

As for your TextCases issue, I bet something strange has happened. NumericArrays kinda changed in 12 and I'm ... not sure what's going on. I would try running SystemOpen@$LocalBase and then clear out Persistence>NetModelIndex

GrammarRules are basically what Wolfram|Alpha parsing was built on. Not exactly that useful for most people. I'm not sure what kind of extension you'd want to see.

Honestly, I expect it not to be updated because actually writing out grammars is such a pain when there's ML methods.

@Searke GeoStyling["ReliefMap"] can take a LightingAngle argument, which is genuinely magical, I love it

@Searke Oh, thanks, I didn't realise. Weird default.

@Searke I ended up deleting the ~/Library/Caches/Wolfram dir and that fixed the issue 🤷‍♀️

there's a bug there. It shouldn't have been possible for that to mess up.

But it's hard to say how.

The default behavior is very pre-internet era

@Searke Well, it's still hard to write something like a chatbot without using GrammarRules, right? I can't arbitrarily say to some ML method "when text like "what is the current time" is entered, run this function"

and have it also match things like "what time is it"

It would be cool if something like that could happen automatically, but the only obvious way right now is to write GrammarRules and think of all the cases manually

In any case, being unable to run GrammarRules as they are now locally is pretty sad :(

Well. I mean this isn't settled at all. There's a lot of value in having a list of well defined grammars that produce well defined behaviors.

@Searke I agree, for sure.

It's mostly a coolness factor thing, I suppose.

But no, in fairness it would be useful for quick prototyping if nothing else.

But it's not unreasonable to say.. have a list of possible actions and then each statement gets categorized into one of those actions and that actions extracts out the parameters that it needs.

a grammar is nice because you could, i guess in theory make recursive statements... but we don't see anyone doing this really with chat bots.

noam chomsky would be very disappointed in the supposed language capabilities of voice assistants.

they aren't parsing. they're associating things statistically.

I actually have no idea if ML is used for google assistant and friends, or if it's just an ever-expanding grammar while they do R&D to move it over to NNs

I suppose in theory this isn't actually that hard in the simplest case. Just build a Classifier with sentence vectors as input...

from everything I've read it's ML

I was just gonna go look at some papers, I wonder what the standard method is

Wolfram|Alpha is very strange in the world of natural language search/commmand engines

But to the point of running GrammarRules locally... that would be nice, but probably a low priority?

I mean you can test most stuff by deploying it and using it and nearly everyone using it at this point has a private cloud instance.

You can use the public cloud too. But I don't know of anyone who has made much of a chat bot that is running it on the public cloud.

@Searke 🤷‍♀️I guess so, I'm just very jealous of my cloud credits. And I live in a tent half the time so there's not much internet sometimes.

@Searke Probably because writing GrammarRules is a bit of a pain ;)

While we're on the subject, being able to add your own natural language control-equals things for your own entities would be cool

interpretations, that's the word I'm looking for

afaik when you do control-equals "minnesota" it goes to WA, which interprets your input query. but if I have my own entity store of, say, popes, I'd like to be able to do control-equals Benedict XVI and have it give me my own entity back

or like, football matches or rock climbers or irish trails, things the built-in knowledgebase will never have

(#knowledgebaseshouldbewikidata)

@CarlLange you can override that :) Or make an equivalent type of cell for calling Wikidata

If you have the function I can show you how to make the cell binding

@b3m2a1 Oh, that sounds cool, I didn't know that was possible

100% possible (if you've sunk enough time into learning how the FE works)

Now that 12 is out I should revisit my wikidata entitystore with all the cool new sparql stuff

and the cool new FilteredEntityClass etc stuff

@Searke quite an interesting paper about what we were talking about: arxiv.org/pdf/1707.05470v2.pdf

@CarlLange The way that Amazon Lex works is that you give it three examples of queries that should trigger a specific action. Lex then uses deep learning to figure out when the user is saying something equivalent to the examples, and extracts the information needed. The key is that you don't have to specify an exact grammar because Lex understands when the intent of the user is the same as in the example queries.