Closed mathematicalmichael closed 5 years ago
six segments.
[x] theory in middle
[x] distribution left
[ ] parameter ID right
[x] plots broken up by input vs output space, both methods on same plots
[x] problem setup at top of examples columns
[x] show equations for the two approaches in the dead middle.
[x] middle column: "A Novel Approach" has DCI formula first, then stats, with labels.
[x] middle column: "what problem are you trying to solve?" and "what is the difference(s) in the two approaches?"
Bayesian Inference updates prior information with a data-likelihood function.
Data-Consistent Inversion updates an initial distribution by inverting a measure constructed from data.
[ ] middle column: model of exponential decay with unknown rate
[x] middle column: bare minimum necessary notation
[x] middle column: references
[x] Title: Using Push-forward and Pull-back Measures for Parameter Identification and Distribution Estimation
I will go ahead and re-structure the document so that we can have a middle-column .tex
file and then four more .tex
files, one for each outer section.
[ ] Bayes/DCI Distribution
[ ] Bayes/DCI Parameter ID
This way, if we decide to swap order (columns have individual methods instead of problem types), we can avoid all copy/pasting.
Okay... my to-do is to re-structure the poster and get the skeleton of it in place.
@yentyu
[x] bolder lines for plots
[x] use colors consistent with other notebook figures
[x] check boxes on github as you go so that we can track things
I did a draft of columns 1 & 2: re-ordering some things and moving figures around.
As a draft, it is more or less finished. Everything that I think should be on the left is there.
@mathematicalmichael: Would you take a look at it sometime on Monday and see if there are any big issues, things that are missing / need to be re-framed or cut?
Otherwise, column 3 and all of the final space adjusting is what is left.
no big issues, just small things
I pushed them to a new branch called edits
for your review (git diff
?) or maybe I'll submit a PR.
Happy with how your plots look but I would consider grabbing the colors used in my notebook so they match on both sides. I dunno.. just my visual aesthetic nagging at me desiring symmetry. Though the colors you chose are readable.
[x] Without moving the location of "Example" on the right, can we somehow make the equations a bit more readable? Font size needs to stay the same, so this is a challenge. [An hour later into learning about latex table formatting and multirow... which I love but have never seen before]: FIXED! Apparently \hline
is a "hack" ... \toprule, \midrule, \bottomrule
exist for this sole purpose. I had no idea.... I've always used \hline
.
[x] Is "Simple Bayes" really the language?
[x] not sure what you meant by needs edit, but I assumed you meant that you wanted the full-version of the images (these just had ten samples for expediency), so thats what I changed with respect to images.
[x] QR code status per #4 ?
Merged your changes. Thanks for them!
Is "Simple Bayes" really the language?
Changed to "Regular Bayes"
Colors
Couldn't use your colors and get the plots to actually not look horrendous. Ack! Too bad I am bad at color palettes. Gonna ignore this, just for time's sake.
not sure what you meant by needs edit, but I assumed you meant that you wanted the full-version of the images (these just had ten samples for expediency), so thats what I changed with respect to images.
Convergence of DCI plot is fine.
For the "Comparison to Bayes" plots, we either need to change title or change concept. I thought we were going for the same type of plots but one column was DCI and other column was Stat Bayes (what I am calling regular Bayes). 2 plots in each column, for one for 10 and one for 100 datapoints (ignore 1 and 50). I also liked your sparse plots.
QR code status per #4 ?
Continuing discussion in #4 .
oh wow I actually didn't notice the "Comparison to Bayes" column heading had changed. I literally just replaced the plots with what was there before. Now I think I understand.
So, I think I understand. Bottom four plots on my end need to be swapped out.
Two of the four remain (D=10 and 100). Then I simply add the posterior*.png
images to the figures. Got it. PR forthcoming.
TITLE: We need to be consistent.
[x] Pushforward and Pullback
[x] Push-Forward and Pull-Back
Does wikipedia have an opinion?
I think the label P.f. update
has enough room to become Pushforward
? Should be "pushforward updated" to be grammatically consistent with "predictive"
I merged and made small edits to everything I feel needs editing.
I feel satisfied with everything: if see anything that needs small edits, feel free to make it happen. Otherwise, I think we're done!
Can I move our names to the bottom? As it is, the title is tooooo squished to the top.
Still have to fix references?
Also I got it working in a binder with pyMC3 now!
And fixed the make
errors.
One reference will be to this repo, another to troy's paper, perhaps another to BET.
I thought you already fixed the references. My bad!
I feel a little uncomfortable moving the names to the bottom only, since I feel like most people will expect to see the names under the title. I am happy for you to make the names small under the title though, if that makes more sense!
Thanks Michael!
-Tian Yu
On Tue, Jul 23, 2019 at 9:17 PM Michael Pilosov notifications@github.com wrote:
Can I move our names to the bottom? As it is, the title is tooooo squished to the top.
Still have to fix references?
Also I got it working in a binder with pyMC3 now!
And fixed the make errors.
One reference will be to this repo, another to troy's paper, perhaps another to BET.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/mathematicalmichael/jsm19/issues/3?email_source=notifications&email_token=AG5YQV7ME7U6HUK2F2EWVZ3QA6USJA5CNFSM4HPR7JZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2U322Q#issuecomment-514440554, or mute the thread https://github.com/notifications/unsubscribe-auth/AG5YQV7JUQQRFFAQ4E4Z2P3QA6USJANCNFSM4HPR7JZA .
hmmm.
well, for some reason the color
thing messed up my compilation, so I removed it and am not a fan of how it looks now. That kind of edit should be made in the sty
file. At the very least we should put our university (maybe CU Denver for space) in the same line as authors.
Feel free to send me a pdf copy if you want a 2nd pair of eyes.
-Tian Yu
On Wed, Jul 24, 2019 at 10:48 AM Michael Pilosov notifications@github.com wrote:
hmmm. well, for some reason the color thing messed up my compilation, so I removed it and am not a fan of how it looks now. That kind of edit should be made in the sty file. At the very least we should put our university (maybe CU Denver for space) in the same line as authors.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/mathematicalmichael/jsm19/issues/3?email_source=notifications&email_token=AG5YQVY26JZNKSNGW6HHBQLQBBTU3A5CNFSM4HPR7JZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2WSPGA#issuecomment-514664344, or mute the thread https://github.com/notifications/unsubscribe-auth/AG5YQV6FCL5UHSGH22T7EALQBBTU3ANCNFSM4HPR7JZA .
I'm figuring out printing, seems like I can just walk up to the printer and do it by USB. Will try today, and worst case I can do it in North classroom.
Cool. Thanks!
wow, caught a lot of spacing things I could improve in < 1 hour jsm19poster.pdf
THE SYMMETRY IN THE POSTER IS SO ON POINT NOW!!!!
thumbs up!
On Wed, Jul 24, 2019 at 5:13 PM Michael Pilosov notifications@github.com wrote:
wow, caught a lot of spacing things I could improve in < 1 hour jsm19poster.pdf https://github.com/mathematicalmichael/jsm19/files/3428778/jsm19poster.pdf
THE SYMMETRY IN THE POSTER IS SO ON POINT NOW!!!!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/mathematicalmichael/jsm19/issues/3?email_source=notifications&email_token=AG5YQV563SDTO6P43R3OLPLQBDAY5A5CNFSM4HPR7JZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2XUG5A#issuecomment-514802548, or mute the thread https://github.com/notifications/unsubscribe-auth/AG5YQV54OKB52FZ5FNKLUG3QBDAY5ANCNFSM4HPR7JZA .
Tian:
[x] Compare this to the poster right now from CSE and decide what content we want to grab from it.
[x] Discuss what we mean by "DCI" because from what I understand, this is theory, which belongs before Parameter identification. I think in general, the whole MT framework is just DCI now instead of "Consistent Bayes."
[x] What are the distribution problems you are thinking about? Are the hierarchical Bayes models where you try to fit sd/mean values for parameters id?