Good Morning!

Mornin' @dharrigan!

I'm so glad Monday and a return to work are coming around -- this has been a brutal week and I'm looking forward to some "normality" or at least "routine" 😐

That's great to hear, I to enjoy the routine/normality of a monday morning and a set of objectives for the week to aim for

It's the reclojure conference this week

so, we're doing some final planning and organising for Friday/Saturday

I did some more podcast interviews over the past couple of days

those should be out soon.

(after I remove my um's and ah's 🙂 )

Oh yeah! Cool! Looking forward to watching the recordings at some point!

Audio only 🙂

But you can imagine it if you wish 🙂

(of the conference talks)

ah! 🙂

I don't listen to podcasts -- can't do just audio -- although the Google Podcasts app on Android, combined with Accessibility that does Live Captioning has finally opened up podcasts to me... so I can READ podcasts!!

ML FTW!

I see the data science workshop stuff being promoted very heavily for re:Clojure...

Yes, it is the theme for this year. Lots of talks and workshops this year. No idea what next year's theme will be.

The recordings will be an education for me -- data science is just not something I'm exposed to much (nor, frankly, have much interest in... but I'll try to make the effort to watch some of it).

It's all a bit magic to me too.

I know someone (from my previous workplace), a data scientist who is now giving Clojure a go for learning and seeing how it'll work for him, instead of Python.

Something else for his toolbelt.

I see travel restrictions are tightening due to omicron -- I'm beginning to wonder what it will all be like when I try to visit mum in March... hoping it will still be possible...

Yeah, I think it's going go to into lockdown again here in the UK.

Masks are being reintroduced in England from Tuesday

Thing is 'tho, I think it's going to be like this every winter time, flu++

I get my booster on Dec 7! We've stayed with masks indoors though (we lifted that very briefly, but I can no longer remember when). I think masks will be the "new normal" and shots/boosters every 6 mos/1 year.

Yeah, I think so too.

My booster is for 21st December.

The "day 2 PCR test" thing makes traveling to the UK "interesting" (esp. with isolation until a negative result) but it's less restrictive than it was last year so we'll see.

The thing the government wants to avoid is the headlines of "Christmas is Cancelled!" that sort of thing

Perhaps after Christmas it'll be more restrictive.

I wouldn't mind "Boris is Cancelled" as a headline 🙂

ha!

I think a lot of people would like that 🙂

Problem it 'tho, they're all just as bad as each other.

At least with Boris, he's, um, entertaining 🙂

Aye, Labour are unelectable again, sadly.

And I suspect for a long long time

The tories are doing policies that are traditionally more labour-esque.

We're living in Upsidedownland 😞

switching the subject, I get the impression that clojure 1.11 won't be far off, esp after the releases of past few days from Alex.

I hope there will be a few more alphas with new stuff in... I'm hoping for functional interface adapters for Java interop!

yes, that would be nice. We're on JDK 17 after all! It's been a while since functional interfaces have been in Java...

Did you get anywhere with your GC selection?

We stayed with G1 because, per my latest blog post, after a few days in production it looks really good.

Vetted, potentially for 1.11: https://clojure.atlassian.net/issues/?filter=10004 Better link: https://clojure.atlassian.net/issues/?filter=10021

https://clojure.atlassian.net/browse/CLJ-2365 is the functional interface ticket

We've had agreat success with the improvements in G1. I'm still keen on re-exploring ZGC (as it has a goal of <1ms for GC's, no matter the heap). I've recently heard about the Shenandoah GC that is supposed to be like ZGC, but with lower requirements for heap sizes.

The story behind ZGC and Shenaandoah is that you'll never have to worry about tuning your GC again once you've switched to either.

Several other tickets are currently targeted for 1.11, including https://clojure.atlassian.net/browse/CLJ-2538 but there's no guarantee that they'll actually make it.

We hardly had to tune G1 so ZGC and onward is pretty appealing 🙂

Adding a tap that returns a value would be super awesome

a small, but handy-dandy thing

I use tap quite a bit, and it would be nice to use it without it returning a nil

I do (doto tap>) right now in thread pipelines but (tap->) would be nicer.

yuuus

I just updated my VS Code/Clover/Portal setup so I can eval blocks (forms) in a regular way (they get tap>'d) with ctrl-; b or I can use ctrl-; shift+t and have them eval'd and tap>'d but appear as tree structures in Portal by default (fully-expanded).

I use tap> so heavily in my daily workflow!

I think portal is a great tool too, defo. I use it often myself. I think I have to get a bigger monitor to have all the windows I need open 🙂

Santa!

I have a 27" and it's "big enough" but I have been lusting after one of those big curved monitors...

yuuuus

me too 🙂

Would you go for a hiDPI display?

I'm getting some monitor envy now,

I don't really know much about hardware. I have seen some reports of problems with hidpi though

Would have thought that some sort of ML would have been interesting for World Singles?

morning

Morning

@slipset <<ML would have been interesting for World Singles?>> What do you mean?

So, please keep in mind that I know nothing about World Singles nor about ML, but here we go: WS is a dating site, presumably they know something about which singles matched up and became an item, and which did not. So given that the amount of singles, items, and not are large enough, you should be able to train a ML thingy on that data, and presto, you’d have a ML model to determine if two people would become an item?

@slipset I’d be surprised if they are not doing this already. I’ve been talking to a few dating startups (they’re still pouring out of the woodwork) and as much as they don’t want an algorithm, they’ll need some at least. I studied the Tinder algo when it was using the Elo scoring system.

I would be surprised if they did since Sean says he knows next to nothing about ML (or datascience) https://clojurians.slack.com/archives/CBJ5CGE0G/p1638167719053900

@slipset The whole matchmaking algo is something I’ve studied for fun.

And Sean being the other backend dev at World Singles.

Ahhh Mr Corfield?

Then my matchmaking algorithm says that you and Sean should have a talk.

Yes, the Corfield.

It would be an honour to have a talk with @seancorfield

And yourself @slipset

I don’t know if World Singles do threesomes 😉

@slipset 😂 too early dude, too early….

sorry 🙂

Well, at least this week got off to a good start 🙂

I’m liking this week already

BTW, my youngest son might be a ML model. He seems to have been trained on music I like (and dislike), so he’s now able with an astonishing level of confidence tell if I like a song or not.

🙂 Yeah they do that. The teen knows the tunes I like.

Interestingly, there is music from 82-86 where the model is off, because there are songs there which normally would fall in the “don’t like” category, but because of sentimental reasons go in the “like” category.

(there might also have been a change in my music taste around that time)

@slipset Any model will have some outliers, at least your not overfitting.

morning

GOOd moRning! :snow_cloud:

This is fun http://maximecb.github.io/Turing-Drawings

Good morning. I’m confusing myself with specs.

morning

gogo normind

(->> "good morning"
     seq
     shuffle
     (apply str))
=> "og onogdimrn"

Something @mpenet said reminded me of this golden nugget of a song. Maybe it’ll brighten your day 🙂 https://www.youtube.com/watch?v=ua1FAlHt_Ys

Good morning, good people!

Morning!

I don't always associate ML and data science, to be honest. re @slipset @jasonbell convo 8-9 hours ago... but I probably should think of them as much closer I guess?

I think I think of them as ML being a subset of data science, but then again, @jasonbell wrote a book about it, so who am I to answer :)

I still associate ML more with the programming language than the number crunching thing

It’s just numbers….. turning things in to numbers, applying numerbery things to more numbers to create numbers.

🙂

But I’ve been spending a lot of time looking at matching making mechanics in general, not just from a ML perspective.

Perhaps a bit to personal, but have you been able to put your research into practical use?

@slipset It was part of a bigger work in progress for DeskHoppa. How do you get the right desk/working space to the right user. You always want the customer to have the best experience. Same applies to StitchFix (Obsessed with that algo), Zara and the dating apps. Once you strip it all back it’s down to who gets what, and why.

Are there algorithms/approaches out there that work when the data set is small or if you lack training data? E.g. in our case (an internal corporate innovation platform) we'd like to show people ideas related to other ideas. But the only think we've found that kinda-somewhat works is using rudimentary text analysis to generate keywords, and use those to trigger plain searches.

And then because the corpora are small and very jargon-heavy the usual NLP models don't give such good results.

might be the sweet spot for Linear Algebra

@orestis In the first instance I would go with @raymcdermott suggestion.

Good to experiment on a number of methods first and see how the results look.