Welcome to the ICM Forum. If you have an account but have trouble logging in, or have other questions, see THIS THREAD.
NOTE: Board emails should be working again. Information on forum upgrade and style issues.
Podcast: Talking Images (Episode 22 released November 17th * EXCLUSIVE * We Are Mentioned in a Book!!! Interview with Mary Guillermin on Rapture, JG & More)
Polls: Favourite Movies (Results), 1998 (Apr 15th), DtC - Ratings (Apr 26th), Coming of Age (Apr 30th)
Challenges: Doubling the Canon, Animation, Middle East
Film of the Week: Foxtrot, May nominations (Apr 30th)

Future polls

500<400, Favourite 1001 movies, Doubling the Canon, Film World Cup and many other votes
Post Reply
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1201

Post by Lonewolf2003 »

TraverseTown wrote: December 6th, 2020, 7:06 am What year will be the first 2020s/current decade poll?
Not at least in 2021, maybe in 2022 (if there’s enough animo to do it than already)

There is a 21st century poll in 2021 and by definition 2020 and 2021 are also part of that century ;) . But very unlikely any will make that list.
User avatar
PeacefulAnarchy
Moderator
Posts: 25938
Joined: May 8th, 2011, 6:00 am
Contact:

#1202

Post by PeacefulAnarchy »

I'm sure a few will make the list.
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1203

Post by Fergenaprido »

I forgot the 2020s would be eligible for the 21st Century poll. :facepalm:

I can think of one or two that would make my list so far.

For a 2020s poll, I suppose we'll vote on it when we're planning the 2022 schedule. If it doesn't make it in that year, I think it's all but guaranteed to happen in 2023.
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1204

Post by Lonewolf2003 »

Fergenaprido wrote: December 6th, 2020, 7:29 pm I forgot the 2020s would be eligible for the 21st Century poll. :facepalm:

I can think of one or two that would make my list so far.

For a 2020s poll, I suppose we'll vote on it when we're planning the 2022 schedule. If it doesn't make it in that year, I think it's all but guaranteed to happen in 2023.
Yes I’m going to give it a spot in the poll to setup the ‘22 schedule (if I’m going to organize that)
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1205

Post by Tim2460 »

Hello There :

Here is the new format i used for the Silent result thread. Silent was maybe not the best way to satart using the posters but anyway it worked not so bad.
https://we.tl/t-NxCgwYFBMj
Fergenaprido & Lonawolf can you take a look at the filmcountV3 sheet to check if i ask for an 250 or 300 films update ?

I have rebuild an result sheet when i removed the 3 Last talking movie and recoutn everything ... you'll see as well the delta between v1 i used for the thread and the recalculated V3
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1206

Post by Tim2460 »

wrong thread ;(
Last edited by Tim2460 on December 7th, 2020, 5:05 pm, edited 1 time in total.
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1207

Post by Tim2460 »

Let me know what 250 or 300 i should send Alisson IYHO for updating the ICM List.
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1208

Post by Fergenaprido »

Tim2460 wrote: December 7th, 2020, 3:55 pm Hello There :

Here is the new format i used for the Silent result thread. Silent was maybe not the best way to satart using the posters but anyway it worked not so bad.
https://we.tl/t-NxCgwYFBMj
Fergenaprido & Lonawolf can you take a look at the filmcountV3 sheet to check if i ask for an 250 or 300 films update ?

I have rebuild an result sheet when i removed the 3 Last talking movie and recoutn everything ... you'll see as well the delta between v1 i used for the thread and the recalculated V3
So is v3 where you reran the script? Or did you just remove those 3 talking films from the results and bump everything else up? The way you've presented it in that tab of the spreadsheet, it looks like v1 is the corrected version, and v3 is the old version, since everything seems to have dropped places in v3, which doesn't make sense.
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1209

Post by Tim2460 »

Hmmm v3 is supposed to be the new version yes...
I'll take a look tomorrow. Too tired right now.
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1210

Post by Tim2460 »

https://www.imdb.com/title/tt0243186/ 0,009467436 1 Perception de Ambiguity (#402) 1120 1123 -3 Cock Fight; No. 2
You're right the head of the collums are wrong : left is last version and right the old i seekv from the old filmcount sorry about that
this way the last film is now position 1120 instead of 1123 on the first version.
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1211

Post by Fergenaprido »

Tim2460 wrote: December 7th, 2020, 8:05 pm https://www.imdb.com/title/tt0243186/ 0,009467436 1 Perception de Ambiguity (#402) 1120 1123 -3 Cock Fight; No. 2
You're right the head of the collums are wrong : left is last version and right the old i seekv from the old filmcount sorry about that
this way the last film is now position 1120 instead of 1123 on the first version.
Okay, so column v1 is the revised version, and v3 is the old version?
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1212

Post by Fergenaprido »

Tim2460 wrote: December 7th, 2020, 7:25 pm Hmmm v3 is supposed to be the new version yes...
I'll take a look tomorrow. Too tired right now.
No worries, take your time.

I also went through the list and compared it with my database, and I found some more films I'm pretty sure are not silent. You may want to add these to your ineligible.txt file before rerunning the script. None of them were near the top 300, but most were in the top 50 of individual ballots, so it may change some of the rankings of Top 300 films overall.

https://www.imdb.com/title/tt0023649/ - Vampyr (1932) - Filmed with the intention of it being a sound film, but produced in three separate languages, so some silent-film elements were used to make production easier.
https://www.imdb.com/title/tt0021577/ - L'âge d'or (1930) - Fully a talkie as far as I can tell.
https://www.imdb.com/title/tt0019644/ - Applause (1929) - Fully a talkie.
https://www.imdb.com/title/tt0018037/ - The Jazz Singer (1927) - The first talkie.
https://www.imdb.com/title/tt0203759/ - L'or des mers (1933) - Filmed as a silent but produced and screened as a talkie, I don't know if this one counts or not.
https://www.imdb.com/title/tt0021273/ - Prix de beauté (1930) - Filmed and screened as a talkie, but a silent version was also made available, so perhaps it should still be allowed.
https://www.imdb.com/title/tt0022694/ - The Blue Light (1932) - Fully a talkie, but apparently can be watched as a silent as well?

As to whether the list should be 250 or 300, my personal preference is 250 because it prevents all 1-vote films from making the list, and only a handful of 2-vote films get it, plus it's in line with our other lists which are usually Top 250s or less. If you want to keep it at 300, though, that's your choice as poll host, and I won't comment on it further. :)

(Your other post came while I was working on this reply)
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1213

Post by Lonewolf2003 »

I will take a look when its recalculated. Take your time TIm, no need to rush. But in general I'm with Ferg about having no 1 vote movies in a list. Plus being it 250 like most decade list is also a nice consistency.
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1214

Post by Fergenaprido »

Found one more talkie.

https://www.imdb.com/title/tt0021375/ - The Donbass Symphony (1930) - early sound film with intentionally little dialogue - first sound film from the Soviet Union by some accounts.

Edit: and another
https://www.imdb.com/title/tt0490566/ - La cartomancienne (1932) - imdb, Light & Cone, and the internet all tell me this is a sound film.
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1215

Post by Tim2460 »

Hello there

Added all but https://www.imdb.com/title/tt0021273/ - Prix de beauté (1930) to the ineligible text file.

Down to 1112 film.

I hadn't check the "low note" thinking they had not a chance to influence the top rankings ... but who knows ; good point Ferge.

Will post the excel with V4 when it's ready
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1216

Post by Tim2460 »

Here's the (final ?) Version

https://we.tl/t-XswpkaH7fs
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1217

Post by Tim2460 »

And the Whole Package, with all the lists for example.

https://we.tl/t-wJkCXIhwyx
User avatar
Fergenaprido
Donator
Posts: 5373
Joined: June 3rd, 2014, 6:00 am
Location: Canada
Contact:

#1218

Post by Fergenaprido »

Thanks Tim.

Everything looks fine as far as I can tell. The only other thing you can do is play around with the half-life if you're unsure of the results. Right now it's set to 30. I don't remember what it was last time, but for most of our polls the half-life is 50, I think, though some have been 20, 25, 30, and others I believe. Lonewolf might know better than me since he hosts more polls than I do. He might be also able to tell what the best half-life is given the average list length for this poll.

With the current half-life, there are 0 one-vote films in the Top 250, and 2 one-vote films in the Top 300.
There are also 6 two-vote films in the Top 250, and 15 two-vote films in the Top 300.
With a Top 250, every eight-vote film makes the list, with a Top 300, every seven-vote film makes the list (the only seven-vote film not in the Top 250 is A Dog's Life at 295), and all but 7 six-vote films make the Top 250 (with 6 of those 7 making the Top 300).
I hadn't check the "low note" thinking they had not a chance to influence the top rankings ... but who knows ; good point Ferge.
Glad you reconsidered this. Just removing a few ineligibles already resulted in a change in the order of the top 5. :)
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1219

Post by Tim2460 »

Indeed : as we say in french "On apprend tous les jours"
--> You learn something new every day ,)
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1220

Post by Tim2460 »

Lone can you take a look at the corrected final list so we can wrap this silent pool?
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1221

Post by Lonewolf2003 »

Tim2460 wrote: December 13th, 2020, 1:24 pm Lone can you take a look at the corrected final list so we can wrap this silent pool?
Sorry been to busy.
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1222

Post by Lonewolf2003 »

Fergenaprido wrote: December 8th, 2020, 9:59 pm Thanks Tim.

Everything looks fine as far as I can tell. The only other thing you can do is play around with the half-life if you're unsure of the results. Right now it's set to 30. I don't remember what it was last time, but for most of our polls the half-life is 50, I think, though some have been 20, 25, 30, and others I believe. Lonewolf might know better than me since he hosts more polls than I do. He might be also able to tell what the best half-life is given the average list length for this poll.

With the current half-life, there are 0 one-vote films in the Top 250, and 2 one-vote films in the Top 300.
There are also 6 two-vote films in the Top 250, and 15 two-vote films in the Top 300.
With a Top 250, every eight-vote film makes the list, with a Top 300, every seven-vote film makes the list (the only seven-vote film not in the Top 250 is A Dog's Life at 295), and all but 7 six-vote films make the Top 250 (with 6 of those 7 making the Top 300).
I hadn't check the "low note" thinking they had not a chance to influence the top rankings ... but who knows ; good point Ferge.
Glad you reconsidered this. Just removing a few ineligibles already resulted in a change in the order of the top 5. :)
I didn't look at the Excel, but going on Ferge's summary I would be for a top 250, cause no 1-votes.

We usually indeed use a half-life of 50. Unless the submitted lists are all very small. I honestly don't see a good reason to have deviated from that based on the list lengths. But I wouldn't change the half-life now again, cause it probably will affect the whole results list even further. Combined with the new list length and changes cause of ineligible titles, and you might as well do the whole results threads again ;)

I would go with a 250 list and post the changes in ranking in the results thread.

But you're the host, you call the shots.
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1223

Post by mjf314 »

Lonewolf2003 wrote: December 15th, 2020, 6:27 pm We usually indeed use a half-life of 50. Unless the submitted lists are all very small. I honestly don't see a good reason to have deviated from that based on the list lengths. But I wouldn't change the half-life now again, cause it probably will affect the whole results list even further. Combined with the new list length and changes cause of ineligible titles, and you might as well do the whole results threads again ;)
The idea behind using different half-lives for different polls is this:

If most people submit lists of about 25 films, they probably like their #1 a lot more than their #25, so #25 shouldn't get a lot of points, and therefore the half-life should be small.

If most people submit lists of 500 films, they probably like their #25 very much, so #25 should get a lot of points, and therefore the half-life should be large.

In other words, the half-life should be roughly proportional to how many eligible films the average voter likes. But like Lonewolf said, avoiding 1-vote films is another thing to consider.

I can see the appeal of using a nice round number, so if you think the ideal half-life is 40 or 60, you might as well go with 50, but for example for the 1001 list, a larger half-life is needed..
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1224

Post by Tim2460 »

Ha that's interesting ... For future polls i need to be able to adjust by myself the Half Life...

Here's the stat of the silent poll :
54 Voters for an average of 81.7 Films lenght list.

Going from a low 10 to ... 413.

There is always an higher number of films on certains Users, PDA being the user with the most vote on all my Poll for ex.

So with the 80 Average ... "ideal" Half-life should/could have been 40 or i got it wrong ?

In other words is the "proportional" .... 1/2 ?
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1225

Post by mjf314 »

Proportional just means that the ratio of half-life to average list size remains the same from poll to poll, but there is no mathematically correct ratio, and different hosts might have a different preference. You can use your own judgment, or you can look at what other polls did to help you decide. You can experiment with different half-lives until you think the results look reasonable. Increasing the half-life will benefit films with many votes. Decreasing the half-life will benefit films with few votes.
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1226

Post by Tim2460 »

Being kind of scientific minded ... tried playing with the half Life a little...

If you wanna play a little with that excel file ... it's very interesting.

https://we.tl/t-OnH4UUfs1b

Collum A to I : notes with 30 / 35 / 40 / 45 / 50 / 55 / 60 / 65 / 70 HL
Collum L to T : Position with 30 / 35 / 40 / 45 / 50 / 55 / 60 / 65 / 70 HL

U to AB : delta of position with the next HL Step (diff wetween 30 & 35, 35 & 40 etc etc)
AC to AK : delta between the HL and the "Usual" HL of 50.

Collum K is the number of votes.

Check the collors down bellow if you wanna see what films change a Lot of note/pos...

I can see a patern below the 300 line where the one vote film got changed a lot more than the 2 & 3 Vote one...

So ok Ferge : it's better to try to avoid including 1 Vote as much as possible theses are too "changing".

If i sort like the result thread By L collum... The River (1928), being the hightest one vote (and a number one vote ,) goes from Pos 290 @ 30 Half life .... to 434 @ 70 Half Life... that a lot of change
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1227

Post by Tim2460 »

For the Top 10.... We don't see any difference between 45 and 60 HL ... So this maybe the "sweet Spot"

Maybe Metropolis got his Victory robbed @ 30 HL

30 35 40 45 50 55 60 65 70
Metropolis (1927) 2 1 1 1 1 1 1 1 1
Sunrise: A Song of Two Humans (1927) 3 2 2 2 2 2 2 2 2
La passion de Jeanne d'Arc (1928) 1 3 3 3 3 3 3 3 3
Sherlock Jr. (1924) 4 4 4 4 4 4 4 4 4
Das Cabinet des Dr. Caligari (1920) 7 7 5 5 5 5 5 5 5
The General (1926) 5 5 6 6 6 6 6 6 7
City Lights (1931) 6 6 7 7 7 7 7 8 8
Nosferatu, eine Symphonie des Grauens (1922) 11 9 9 8 8 8 8 7 6
Chelovek s kino-apparatom (1929) 8 8 8 9 9 9 9 9 9
Körkarlen (1921) 10 10 10 10 10 10 10 10 10
Bronenosets Potemkin (1925) 9 11 11 11 11 11 11 11 11
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1228

Post by Tim2460 »

Image

Hard for me to say witch HL is "better" ...
User avatar
PeacefulAnarchy
Moderator
Posts: 25938
Joined: May 8th, 2011, 6:00 am
Contact:

#1229

Post by PeacefulAnarchy »

I think looking at the results as films instead of vote distributions is a bad way to judge. I use median list length as a proxy, but if you want to judge by film placement you should be looking at something like: how many top 10/25/50/100/total votes films got. Higher half-lives will minimize the effect of top ranking, while smaller half lives emphasize the importance of a top ranking.

I use list length as a proxy because it balances the idea that most people's lists should have reasonable distribution throughout their list. i.e. if a lot of lists have everything with a lot of points (lets say over 10) then the half life is too high, if a lot of lists have many films under 1 point then the half life might be too low. Of course some lists will have both of these properties, but the bulk shouldn't. I say "proxy" because list length distribution is not the same for all polls, so sometimes you want to adjust one way or another.


Edit: I do wonder if we're ready to move on to something even more complex than exponential decay, but when I floated the idea last year the response wasn't too enthusiastic, and the few discussions since then have been split between people wanting more emphasis on the top of their lists and people wanting more emphasis on simply placing on a list.

A huge work involved idea would be to allow people to choose a distribution, but a) some distribution would still need to be the base to make things equitable, and I can't imagine a way of doing it outside of a webform which is a ton of work.
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1230

Post by mjf314 »

I think asking people to choose their own distribution would too much work for the participants.

I don't remember the discussion, so I'm not sure if you had any other suggestions.
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1231

Post by Lonewolf2003 »

We never had a HL above 50, so I recommend not using any higher for this one either.

Cause a few very large or small lists can indeed skew the average list size very much, I also always look at the median. Both literally and more just looking at how long most lists are approximately (f.e. are most list about 20 or 100 titles long). I usually also play around a bit just so we have a nice round list length without any 1 or 2 votes movies (or just a few).

The minimum amount of votes I find acceptable to be in the results depends on the amount of voters. It’s makes more senses if there are 2 votes films in a list when only 20 people submitted a list than when 100 people did.
User avatar
Lonewolf2003
Donator
Posts: 11144
Joined: December 29th, 2012, 7:00 am
Contact:

#1232

Post by Lonewolf2003 »

Getting back on topic on which HL to use for the Silents poll: I recommend either sticking with 30 or going for 50.

I would have gone with a HL of 50, cause that's the standard and there's not much reason to change it for this poll based on the list lengths. Plus it's also what was used last time. The info on the OP of the last nominations thread isn't very clear, but I we didn't use any lower HLs than yet (as far as I remember). On top of that with a HL of 50the list length can still be 300, cause there are no 1-votes in the top 300 then. Or only one 2 vote in the top 250.
That the #1 is different only for 30, but the same for 35 or above, for me would be a reason not to go with 30

Looking at the file, I see that especially lower in the list the rankings change considerably when comparing 30 to 50. I doubt now preferable it is to post a list with such different rankings than the Result thread. No. That would be the main reason to stick to 30.

We never had a HL above 50, so would never use that for this. And using any between 30 and 50 just seems like a half-way solutions between those two options to me.

My preference would be going with top 250/300 with a HL of 50. And to post the whole list in the Results Thread with the changes of rankings between the new and old HL, including the movies that fell of the top 250/300.
Like this f.e.
#1 Metropolis (1927)(#2, ^1)
#2 Sunrise: A Song of Two Humans (1927) (#3 ^2)
#3 La passion de Jeanne d'Arc (1928) (#1 downward arrow 2)
User avatar
Tim2460
Posts: 6036
Joined: October 1st, 2018, 7:23 pm
Location: Dijon, France
Contact:

#1233

Post by Tim2460 »

Just read the Old Nomination thread again and the most important info i found was, By PA :

viewtopic.php?f=6&t=1489&start=80#p216284

If i quote
Based on the list lengths I'm seeing (average length seems to be in the 70-80 range), my previous suggestion should be about right:
=100*(0.5)^((R-1)/40)
where R is the cell with the ranking
We are cleary on the same Average ...

Was the 40 an proposition for HL = 40, seeing the average lenght of the lists so far ?
User avatar
pitchorneirda
Posts: 796
Joined: February 11th, 2019, 12:07 pm
Location: France
Contact:

#1234

Post by pitchorneirda »

Median length of a list submitted to the Silent poll: 61.5
Mean length of a list submitted to the Silent poll: 81.8

Median length of a list submitted to the 60s poll: 130.5
Mean length of a list submitted to the 60s poll: 145.5

Long story short, it would be weird to me to use the same half-life for both polls, especially given that the silent one covers roughly 3 decades vs 1 for the 60s (i.e. people have clearly less true favourites in the silents and some of the voters had trouble filling up a decent list)

More consistency between the results for half-life ranging from 35 to 50 doesn't necessarily mean more relevancy
"Art is like a fire, it is born from the very thing it burns" - Jean-Luc Godard
User avatar
PeacefulAnarchy
Moderator
Posts: 25938
Joined: May 8th, 2011, 6:00 am
Contact:

#1235

Post by PeacefulAnarchy »

mjf314 wrote: December 16th, 2020, 10:46 am I think asking people to choose their own distribution would too much work for the participants.

I don't remember the discussion, so I'm not sure if you had any other suggestions.
A suggestion I floated was exponential + constant. It would have to be something small enough to not overshadow the exponential value, but big enough to be meaningful, a way of boosting the tail a bit so that the # of voters has a bit more weight. Exponential plus linear also might work in a slightly weaker way. I'm not sure I support these, I just thought they were interesting ideas.

Choosing the distribution wouldn't have to be a lot of work on participants if done right. Like choosing ranked vs unranked vs partially ranked etc. We choose a default distribution for those who don't want to choose, offer a few standard distributions as options, maybe offer a free choice distribution if anyone wants. From the participant side it's no more than 10 seconds unless they feel so strongly they want to make their own distribution. It's the actual implementation of a) communication of that preference in a standardized way and b) the poll compiler having to actually implement it, that would be a pain in the ass. It could only work with a web form that lets people upload a list and choose from a drop down box or something and that's a lot of prework for something I'm not sure is worthwhile. But if it existed list submission wouldn't be any more work for participants.
User avatar
PeacefulAnarchy
Moderator
Posts: 25938
Joined: May 8th, 2011, 6:00 am
Contact:

#1236

Post by PeacefulAnarchy »

Reading a bit more of this discussion I want to add that consideration of where 1-vote/2-vote films end up should have no bearing on half life choice. It should probably have a bearing on list length choice after the half life is chosen, but not the reverse. A list length should be chosen to only display meaningful results, choosing a half life based on what lets you make a longer list just risks choosing a half life that is too high.

I'm not sure anyone was saying otherwise, but the discussion seemed unclear to me.
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1237

Post by mjf314 »

I would be ok with exponential + constant, but I'm not sure if I prefer it to exponential by itself.

For example, if someone decides to submit a list of their top 1000 silent films, they probably think the #1000 film is just average. Does it really deserve a significant amount of points?

For the choose-your-own-distribution idea, I guess I'd have to see the implementation before I can judge it.
User avatar
PeacefulAnarchy
Moderator
Posts: 25938
Joined: May 8th, 2011, 6:00 am
Contact:

#1238

Post by PeacefulAnarchy »

mjf314 wrote: December 16th, 2020, 11:17 pm I would be ok with exponential + constant, but I'm not sure if I prefer it to exponential by itself.

For example, if someone decides to submit a list of their top 1000 silent films, they probably think the #1000 film is just average. Does it really deserve a significant amount of points?
Yeah, if we were to implement it it would require a good deal of thought and discussion to determine what a reasonable constant would be. The assumption would be that being on your list means it isn't just average but that a submitted list is curated.

The +(other function related to rank) would take care of that issue, but each choice of another function in there would have different effects and settling an ideal one is non-trivial.
For example 100*(halflife+ 0.5/(x+5)) adjusted so that number 1 is still 100 points is pretty similar to exponential decay but boosts the tail a bit. But is it worth the difference to complicate the formula? Should it be x+10 instead of x+5? 0.7 instead of 0.5? There are a lot of subtle questions. I think a function like that would be an improvement if chosen correctly. I'm not sure the level of improvement is worth the complication for our purposes.
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1239

Post by mjf314 »

Here's another possible option:

Image

Blue = 100 / 2 ^ ((rank-1) / 50)
Red = average(blue, 900 / ((rank+26) ^ (2/3)))

Blue is the standard exponential decay with half-life = 50.
In the red line, #500 gets 6.955437 points, and #1000 gets 4.4237 points.

The formula is a little bit complicated, but maybe someone can come up with a simpler formula that has a similar shape.
PeacefulAnarchy wrote: December 16th, 2020, 11:36 pm For example 100*(halflife+ 0.5/(x+5)) adjusted so that number 1 is still 100 points is pretty similar to exponential decay but boosts the tail a bit. But is it worth the difference to complicate the formula? Should it be x+10 instead of x+5? 0.7 instead of 0.5? There are a lot of subtle questions. I think a function like that would be an improvement if chosen correctly. I'm not sure the level of improvement is worth the complication for our purposes.
Is there a typo in this formula? When I use a half-life of 50, it seems to give more than 5000 points to everything.
mjf314
Moderator
Posts: 11807
Joined: May 8th, 2011, 6:00 am
Contact:

#1240

Post by mjf314 »

Here's a simpler formula (yellow):

Image

Blue = standard exponential decay
Red = the formula in my previous post
Yellow = 100 * 50 / (rank+49)
Post Reply