<scribe> ScribeNick: myles
Garret: Updated Simulation
Results
... We have results!
... New data set has more sequences
... We're still sampling the Latin/Cyrillic/Greek set just
because it would take too long to run if we didn't
... But for CJK, we get more sequences
... Got set up to simulate range requests
<JHudson> Can't remember zoom password
Garret: Combined multiple patch
subset settings into a single number
... We also dropped hints and variable data from all fonts in
the data set
chris: I understand why you did this in terms of doing a fair comparison. This isn't because we can't ever do hints and variable data, it's just "for now" right?
Garret: Yes. If we think it's important to include this stuff, we can update the optimizer. But I don't think it will change the results
chris: We don't want to say "you can't use variable fonts" with our new technology.
Garret: Yeah.
... New approach to digest results down into a smaller set of
numbers.
<Vlad> * to all participants: don't forget to present+ yourself for attendance record
Garret: Using these percentiles
to create a cost distribution
... We normalize the numbers by the unicode range method, on a
per-sequence basis
... <shows graph>
... The dot in the middle is the median, the ends of the bars
is the 5th and 95th percentile
... <another graph>
... <another graph>
... I'm going to send out the numbers that these graphs were
based on in the next couple of days
... This is a summary
Vlad: Can you share the presentation too?
Garret: Should we use this new summary method going forward?
chris: Normalizing in terms of what's the best current thing is correct
Vlad: People who are not familiar with our work would have a reference point
jpamental: I know it's possible to do this, but how many people are even getting the benefit of that?
Garret: I went back on forth on whether I should use whole font or uncode range for the basis. Self-hosters use whole font
myles: so maybe we want both
Garret: absolutely
chris: Also, we want wholefont vs
unicode range
... A lot of people host their own fonts, plop it on the
server, and off they go. They are leaving perf on the table
jpamental: A lot of what people talk about for perf is for self-hosting because they feel they get the best perf that way. We want to say "yes, you are optimizign things for yourself, but you can get more by doing X"
Garret: We could just do three sets of graphs
Vlad: wholefont vs unicode-range would explain why these are used for baselines
Garret: We could report more percentiles on these intervals to get more data about the shape
myles / Garrett / chris: We can use colors or opacity to get a better idea of what the curve shape looks like
s/Garett/Garret/
Garret: Will try to send out new
data set in the next few days.
... along with font library and optimized font library, and
another pull request.
... Then we'll be all set up and ready to go.
Vlad: We have 4 open action
items.
... One action item is on Garret to post data set.
... The other three, analysis framework, integrated approach,
and proposal to evaluate perf numbers, can be
closed.
<scribe> ACTION: 215 close
<trackbot> Error finding '215'. You can review and register nicknames at <https://www.w3.org/Fonts/WG/track/users>.
<scribe> ACTION: 223 close
<trackbot> Error finding '223'. You can review and register nicknames at <https://www.w3.org/Fonts/WG/track/users>.
<scribe> ACTION: 224 close
<trackbot> Error finding '224'. You can review and register nicknames at <https://www.w3.org/Fonts/WG/track/users>.
close action 215
<chris> close ACTION-215
<trackbot> Closed ACTION-215.
close action 223
close action 224
<chris> close ACTION-223
<trackbot> Closed ACTION-223.
<chris> close ACTION-224
<trackbot> Closed ACTION-224.
<Vlad> https://w3c.github.io/PFE-analysis/report/evaluation-report.html
Vlad: The next one is about the evaluation report with the link to the draft he prepared. Looks great. Sets up the stage for the introduction and presentation. Chris, please introduce it.
chris: I was concentrating on the
intended audience who is not aware of fonts. One audience is
the AC, another is people who are interested but don't know
details. I started explaining things from scratch. I introduced
current work about pros/cons of current work.
... A lot of the network type explanations are not final, just
throwing together
... Pointed out this is a simulation, not a "real"
network.
... I didn't understand the webpage corpus, which methods that
affected
... in out of scope, it says we're not creating a wire
protocols
... analysis and conclusion are empty because we haven't got it
yet
... tables are good but charts are better for our intended
audience
... I was thinking that actually I'm assumign knowledge of
OpenType and what an OT table is, and how much data in
different tables it takes to render one glyph in arabic, so I
might add a section that explains that with diagrams, just so
people can see the data you need is scattered throughout
... This explains about for byte ranges, you're only subsetting
outlines.
... I'm welcoming of criticism. Please read!
Vlad: With this draft, integrated
with final results, I want to discuss the tentative
timeline.
... Garret, you mentioned with all these updates, you can get
the final sets of numbers and illustrations within the next
couple of days?
Garret: Let's say by the end of the week
<chris> AC meeting (Member only) https://www.w3.org/2020/10/TPAC/ac-agenda.html
Garret: I need to merge code,
publish data set, and explore suggestsion for improving
graphs
... and sending out results
Vlad: okay, one more week to get
everything ready.
... And an extra week to integrate it into the evaluation
report.
... That brings us to Sept. 28.
... At that point we can finalize our recommendation and
decisions.
... Is this timeline okay?
Garret: Works for me
chris: works for me
Vlad: let's aim for Sept. 28 to
be the final deadline.
... chris, how long would publication take?
chris: the editor's draft is immediately available, and TR the next tuesday or thursday after we decide it's good.
Vlad: So Oct 1 or Oct 6
chris: yep
Vlad: Which is good timing for TPAC
chris: There's a link above about when the AC part of TPAC is. Are we going to present to the AC?
Vlad: Our charter has a
deliverable there. The reason our charter was approved for this
long was in 2020 we will deliver an updated communication to
AC.
... That was the reason why I'm including the AC presentation
as a deliverable.
... It can be a presentation or a communication of a
link.
... We can make a decision as we go along.
chris: The live session to the AC is on the 20th of october.
Vlad: Our communication should be available a week before that.
chris: If you're speaking, the 21st of September is the deadline for slides.
Vlad: I don't think this requires a presentation.
chris: I agree.
Vlad: We just want to show good
results and tell them we're on target.
... As for TPAC: do people have opinions?
... We don't need an official meeting during TPAC. We can just
have a regular conf call that week.
... We can start planning for the future and next steps.
... Are there opinions?
jpamental: Is it worth it for us to publicize that meeting as a "TPAC" meeting and invite people to sit-in?
Vlad: The typical TPAC meeting
isn't much different than our normal ones. Outside people can
attend with chair permission. This group operates in public.
The only obstacle to add attendance is the password-pretected
zoom room. But this is just for security, not to keep people
out. We can definitely do something like that invite people
in.
... That would be a question for chris. How easy is it to set
up a separate zoom room?
chris: If we want to do a public consultation, it's a good idea, we would do it as a separate booking. The password can be shared. It's a good opportunity to invite people to ask us questions.
Vlad: We can even re-use the same
timeslot.
... Okay, we'll do that.
... We should aim for the second part of TPAC, after our AC
report is already published.
... That brings us to October 26.
... If there is a need for joint meeting presentations and
lightning talks, I'm not sure.
... Because time is valuable. Virtual meetings have limited
timespan.
... chris, what do you think?
chris: I think it's enough to just make an announcement and take any questions during the live session if needed.
Vlad: Would those live sessions limited to 1.5 hours like they were in the past?
chris: I don't know.
Vlad: 1.5 hours is a reasonable time limit.
myles: So our call on Oct 26th will be 90 minutes long?
Vlad: No, just one hour.
<chris> fonts and text CG call on 21 Sept partly overlaps this call
<chris> first call for this new CG
<chris> so skip our WG call, Next call on 28th
<Vlad> I guess we need to present+ all attendees
This is scribe.perl Revision of Date Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00) Succeeded: s/We/Garret: We/ FAILED: s/Garett/Garret/ Succeeded: s/show/tell/ Default Present: Vlad, sergeym, myles, Garret, jpamental, JHudson, Ned, ChrisL Present: Vlad sergeym myles Garret jpamental JHudson Ned ChrisL Found ScribeNick: myles Inferring Scribes: myles Found Date: 14 Sep 2020 People with action items: 215 223 224 WARNING: Input appears to use implicit continuation lines. You may need the "-implicitContinuations" option. WARNING: IRC log location not specified! (You can ignore this warning if you do not want the generated minutes to contain a link to the original IRC log.)[End of scribe.perl diagnostic output]