"Should I Keep This Riven?" and "What Traits Do I Want"? Answered for Every Single Weapon and Trait (Trading)

[-ChaosNyx-]

1 hour ago, Voltage said:

This is a noble cause, but the thing you need to wrestle with is that auctions are all the Rivens that haven't sold. There is also no time limit on auctions, so players will just wait things out until the price reaches that predetermined price they had in their head that they were too insecure listing it for originally. There are also "fake" listings where players list their collection for a weird value and then some nonsense like "124356 = not for sale" on their profile. Lastly, Riven Mods will remain speculative as long as the nature of their randomization stays the same.

Finding out how optimal Riven value is worth is moreso being "in the know" with the Riven game. What I mean by this is being connected to other people so that before you list something, you get some speculative feedback before throwing your dart on the board for others to see. There are also the "Unrolleds" and "Grades" of Rivens to consider when pricing up stats. If you want to know whether the weapon you rolled the stats for is good, you should do your homework on what is the current meta, upcoming Prime Access, recently released content, or leaks for augments/weapons. What drives up price from a select weapon is because the item is new, established, or upcoming and yet to be revealed.

This chart is just a bunch of numbers, and it's not useless information, but it's not helpful for the kinds of questions you aim to answer. Riven Mods and their value are based mostly on "feel", and that's because of how the community has driven prices on things. The only thing datasets can show players for Riven pricing is what weapons are worth flipping unrolleds with to be quite honest.

And again, I think your posts as of late are interesting, but this approach doesn't address broader questions or issues players have with Riven Trading.

Incredibly grateful for the in-depth feedback and the time you took! You raised some very valid points.

For time limits, fake listings, absurd values, and outliers, I put methods in place both before and during training to remove them. Any bad data that remains, likely only makes up a small portion of the total. Regarding the data being based on listings rather than actual trades, I completely agree this is a limitation—I badly wanted actual trade data, but this was the best I could find. Future extensions of the project will aim to correct for the expected bias in listings data (using developer summary stats as one approach). While this would shift the price distribution, the relative impact of traits should remain largely unaffected.

I also agree that the market is speculative and will remain so. However, that doesn’t mean we can’t gain useful insights from learning how people tend to value Rivens. These insights can then be factored into personal decision-making.

The model aims to be like your “guy in the know.” The speculative feedback you get from people can be thought of as models themselves—trained on previous trades and bringing their experience to the table. But imagine you had a friend who had personally seen 200,000 listings just in the last week. Wouldn't you think he probably has the best current insight? Each of the listings he saw was created by players who try to price in factors like upcoming news, meta shifts, rolls, grades, etc.—some overvaluing, some undervaluing.

Market movements, such as those that drive up prices, would be reflected in the data the model is trained on. So, indirectly, the model is learning about recent behaviour.

The chart comes from this model, which is trained on data that incorporates many of the concerns you mentioned. It’s not perfect—no model can be. But I believe it’s premature to write off something just because it’s not flawless. I also respectfully disagree that the chart doesn’t address a large problem players have with Rivens. The chart provides two key insights: (1) the expected value and ranking of every weapon, and (2) the relative impact of traits for specific weapons. It doesn’t attempt to answer the exact value of a Riven—that’s the model’s job in the project.

I would really look forward to any more feedback!

 

[-ChaosNyx-]

1 hour ago, Karyst said:

Price check https://imgur.com/a/O1VscPi

https://imgur.com/a/SuWaFoy

Reminder: The stats are not gospel truth, but instead insight from a model looking at listings.

Cool riven though regardless :P

[kulfii]

the passion is irradiating 

a kuva survival run couldve been nice fr

[dwqrf]

il y a une heure, -ChaosNyx- a dit :

The model aims to be like your “guy in the know.” The speculative feedback you get from people can be thought of as models themselves—trained on previous trades and bringing their experience to the table. But imagine you had a friend who had personally seen 200,000 listings just in the last week. Wouldn't you think he probably has the best current insight? Each of the listings he saw was created by players who try to price in factors like upcoming news, meta shifts, rolls, grades, etc.—some overvaluing, some undervaluing.

Unless you know, too many people start over-using your tool and it start becoming the dominant pricing method, leading to people flipping riven doing wrong pricing as your machine will start using its own prices as data and start exibit more and more flaws. Ever thought about that loop ? 🔄

[Karyst]

4 hours ago, -ChaosNyx- said:

https://imgur.com/a/SuWaFoy

Reminder: The stats are not gospel truth, but instead insight from a model looking at listings.

Cool riven though regardless :P

Cool thanks. Although do explain how exactly is the chance for better calculated please.

[Karyst]

7 hours ago, Zakkhar said:

Not really. People in general have no idea about rivens and just pretend to know or take someone else word for it. The aim of the auctions is not to sell best rivens, but to make most plat on them. That is why shiny stats like crit and %dmg are so popular. 

I'd argue that that is no longer the case, mostly. We often see newer players asking for a price check in-game and are often guided to various riven focused sites

[Karyst]

7 hours ago, -ChaosNyx- said:

I'd say to some extent yes. However, in trading, other factors like player enjoyment, hype, and such can play a massive role. So this is why when it came to predicting value in trading, I used actual listed trades rather than any more 'objective' measurement such as gun performance.

Was this comment made under the knowledge the spreadsheet is targeted at trading and not gun performance?

You can suddenly receive an info on new incarnon dropping and then a riven that was let's say 10p lowest unrolled goes to 300p lowest unrolled. How does your program handle that?

[sly_squash]

I like this OP.  :>

7 hours ago, Voltage said:

This is a noble cause, but the thing you need to wrestle with is that auctions are all the Rivens that haven't sold. There is also no time limit on auctions, so players will just wait things out until the price reaches that predetermined price they had in their head that they were too insecure listing it for originally. There are also "fake" listings where players list their collection for a weird value and then some nonsense like "124356 = not for sale" on their profile.

Yeah, this is an excellent point and why I'm leery of pulling in data from listings that haven't sold in months, years, or aren't even intended to sell.

For what it's worth, I'm one of the top sellers in the riven market.  According to riven.market I'm approaching 5k rolled riven sales from that site.

To confront the problem that the information available on market sites generally refers to rivens that have not sold, I began logging the full details of my own transactions of riven configurations and prices that that did sell.  My last post regarding this was roughly a year ago and I had about the full details of my previous 800 or so rivens that sold.  At this point, my log is approaching the full details of my last 2,000 riven sales, so I'll likely be making an updated thread soon.  This might be some good data for you to mine for your model. 

Here's my previous thread from last year.

 

Edited September 20 by sly_squash

[-ChaosNyx-]

11 hours ago, dwqrf said:

Unless you know, too many people start over-using your tool and it start becoming the dominant pricing method, leading to people flipping riven doing wrong pricing as your machine will start using its own prices as data and start exibit more and more flaws. Ever thought about that loop ? 🔄

Yes! Fortunately, I don't believe the raw pricing tool will ever gain enough popularity to significantly impact the market. I expect only the spreadsheet, which is just a general guide, might reach a broader audience, and even then, its influence would be limited.

However, if it did gain traction, you're right that there wouldn't be a clear solution. You'd have people identifying weaknesses in the model, some betting against it, others ignoring it altogether and returning to a more traditional speculative market, and so on.

[-ChaosNyx-]

8 hours ago, Karyst said:

Cool thanks. Although do explain how exactly is the chance for better calculated please.

The model is applied to every possible riven permutation of that weapon, weighted by their probability according to how rivens are randomly generated. So with every possible riven having its predicted price, we can then compare your current riven against that distribution to identify where it ranks. So of all 100% of rivens that could be generated on a re-roll, your riven has a higher predicted price than 98.4% of them.

[-ChaosNyx-]

7 hours ago, Karyst said:

You can suddenly receive an info on new incarnon dropping and then a riven that was let's say 10p lowest unrolled goes to 300p lowest unrolled. How does your program handle that?

The tool's sensitivity to immediate events is really up to the user. It's highly customizable in that way. For example, if you want to only include data from the past day, you can filter for that when analysing the market. In this case, the model would be highly sensitive and reflect real-time changes.

Ultimately, the tool isn't magic—it's just a reflection of the market. If you want that reflection to capture general value, you can broaden the time span significantly. If you want it to focus less on general trends and more on specific factors like hype or other short-term influences, you can train it on a shorter time span.

[-ChaosNyx-]

7 hours ago, sly_squash said:

I like this OP.  :>

Yeah, this is an excellent point and why I'm leery of pulling in data from listings that haven't sold in months, years, or aren't even intended to sell.

For what it's worth, I'm one of the top sellers in the riven market.  According to riven.market I'm approaching 5k rolled riven sales from that site.

To confront the problem that the information available on market sites generally refers to rivens that have not sold, I began logging the full details of my own transactions of riven configurations and prices that that did sell.  My last post regarding this was roughly a year ago and I had about the full details of my previous 800 or so rivens that sold.  At this point, my log is approaching the full details of my last 2,000 riven sales, so I'll likely be making an updated thread soon.  This might be some good data for you to mine for your model. 

Here's my previous thread from last year.

 

Now this is awesome! Since the beginning, I really wanted to get my hands on a lot of actual trade data. While I don't think I can just chuck this in with the other data, nor create an entirely solitary model from this data, it still undoubtably can provide some valuable insight. I'm thinking this will go a long way in the project's extension of introducing a bias adjustor, in which the goal is to try to map the predicted listing prices to trade prices.

Ngl I'm a little burnt out from doing Kuva maths and ML for my space ninja wizard nuke game, so I'll probably not be doing this anytime soon haha. However, if I get back to it, I'll definitely come back here. Also, I'd try to contact you to let you know what I intend to do with the data to make sure you're okay with that :)

Oh, and on your point on the eeriness of the data, I believe I replied to that poster's original comment to address that, if you're interested.

[Karyst]

7 hours ago, -ChaosNyx- said:

The model is applied to every possible riven permutation of that weapon, weighted by their probability according to how rivens are randomly generated. So with every possible riven having its predicted price, we can then compare your current riven against that distribution to identify where it ranks. So of all 100% of rivens that could be generated on a re-roll, your riven has a higher predicted price than 98.4% of them.

I don't understand. I mean how does it calculate the possibilty for better roll, that's what I meant. Nothing to do with price. 

[Karyst]

1 minute ago, Karyst said:

I don't understand. I mean how does it calculate the possibilty for better roll, that's what I meant. Nothing to do with price. 

Probability to roll a better riven in the case of riven I linked is 1.6%. How is that calculated and how can it be so low considering it's possibly the worst roll you can get?

[-ChaosNyx-]

2 hours ago, Karyst said:

I don't understand. I mean how does it calculate the possibilty for better roll, that's what I meant. Nothing to do with price. 

Well, “better roll” is in terms of price. So when the model says “there is a 1.6% chance of getting a better roll”, it is saying that of all the possible rivens you could get for this weapon, only 1.6% of them have a higher price.

[-ChaosNyx-]

2 hours ago, Karyst said:

Probability to roll a better riven in the case of riven I linked is 1.6%. How is that calculated and how can it be so low considering it's possibly the worst roll you can get?

So that's the surprising part! The predictions can go against our intuition. But this can probably be explained by the model seeing far more data than you ever will. It has seen that generally, your riven and its traits would “fit” best into the market if it had that predicted price. At the end of the day, it's not divine knowledge. It's a tool that gives you a suggestion based on what it's seen. You get to be the final say.

[Karyst]

7 hours ago, -ChaosNyx- said:

So that's the surprising part! The predictions can go against our intuition. But this can probably be explained by the model seeing far more data than you ever will. It has seen that generally, your riven and its traits would “fit” best into the market if it had that predicted price. At the end of the day, it's not divine knowledge. It's a tool that gives you a suggestion based on what it's seen. You get to be the final say.

I understand now, thank you.

[Karyst]

17 hours ago, -ChaosNyx- said:

The tool's sensitivity to immediate events is really up to the user. It's highly customizable in that way. For example, if you want to only include data from the past day, you can filter for that when analysing the market. In this case, the model would be highly sensitive and reflect real-time changes.

Ultimately, the tool isn't magic—it's just a reflection of the market. If you want that reflection to capture general value, you can broaden the time span significantly. If you want it to focus less on general trends and more on specific factors like hype or other short-term influences, you can train it on a shorter time span.

This sounds very much like fragrantica pulse.

[-ChaosNyx-]

6 hours ago, Karyst said:

This sounds very much like fragrantica pulse.

#notaffiliated

[Dunkelheit]

Am 19.9.2024 um 22:57 schrieb Zakkhar:

Nobodyy is hating on the person. All am saying is that it doesnt reflect actual useful stats and most people who could  use it, wont.

Dude, calling someone elses work "garbage" is the very definition on hating on a person. Just be friendlier next time or keep it to yourself.

[Dunkelheit]

Am 19.9.2024 um 23:01 schrieb Zakkhar:

Not hating and not on the author. There is nothing to talk, I see what he wrote, I can see the stats resulted from that. I can judge for myself checking some popular weapons and the listed stats prio for them, considering I happen to know the meta builds.

Also It (at least i did not notice) doesnt list most iomportant thing, of how good/popular is the weapon, what is the current dispo, which strongly affect the riven price. You can get a god roll for niche weapon/with low dispo and it will still be harder and cheaper to sell than any roll for meta weapon/high dispo. 

There are 2 ways to respond to a posting:
1. Your opinion/work/effort/result is dumb/garbage/troll/overpriced

2. This is my opinion and here are my arguments.

It IS really possible (I swear) to stay friendly and respectful and still voice an opinion that is contradicting the OPs opinion.

[Zakkhar]

6 minutes ago, Dunkelheit said:

On 2024-09-19 at 10:57 PM, Zakkhar said:

Nobodyy is hating on the person. All am saying is that it doesnt reflect actual useful stats and most people who could  use it, wont.

Dude, calling someone elses work "garbage" is the very definition on hating on a person. Just be friendlier next time or keep it to yourself.

Context is everything. This is my initial post in this topic:

The post you are refering to was a response to someone else asking how it is made, who assumed it was made certain way and since it is clearly not that was the shortest and most accurate describtion in this ver context and meaning.

 

And yes, he understood it correctly:

 

6 minutes ago, Dunkelheit said:

It IS really possible (I swear) to stay friendly and respectful and still voice an opinion that is contradicting the OPs opinion.

Read the topic, I was not adressing OP in any means. 

[-ChaosNyx-]

Updated spreadsheet to include some extra stats.

[-ChaosNyx-]

 

Don't mind little ol' me just posting this here so I can see it in my steam overlay, lol.

https://i.imgur.com/bGnD3Sb.mp4

Edited September 27 by -ChaosNyx-
animation

[-ChaosNyx-]

Gave the spreadsheet a face lift as well as including even more metrics :P