now i can finally have a lore-accurate Defect

Good job nice

I love this mod but it has so many mod conflicts, even when i only use QoL mods and not mods that add content :( Really this is my favourite mod but I can't use it half the time!

This Mod seems to be broken atm as it crashes constantly.

Why does the AI love clash

This mod crashed game...

Game crashed.
Version Info:
- Java version (1.8.0_144)
- Slay the Spire (03-07-2022)
- ModTheSpire (3.29.0)
Mod list:
- basemod (5.44.1)
- ArchipelagoMW (0.1.2)
- block-reminder (1.2.0)
- coloredmap (2.4.1)
- GooglyMod (1.9.0)
- HighlightPath (0.0.3)
- intentgraph (0.2.1)
- mintyspire (2.5.5)
- relicsorter (2.0.0)
- FightPredictor (1.0.4)

Mods in stacktrace:
- FightPredictor (1.0.4)
Cause:
java.lang.NullPointerException
at FightPredictor.util.StatEvaluation.getWeightedAvg(StatEvaluation.java:88)
at FightPredictor.CardEvaluationData.createByFunction(CardEvaluationData.java:106)
at FightPredictor.CardEvaluationData.createByAdding(CardEvaluationData.java:58)
at FightPredictor.patches.com.megacrit.cardcrawl.screens.CombatRewardScreen.CombatRewardScreenPatches$EvaluateCardRewards.Postfix(CombatRewardScreenPatches.java:26)
at com.megacrit.cardcrawl.screens.CombatRewardScreen.setupItemReward(CombatRewardScreen.java:108)

UI meaning isn't obvious when removing a card.

- Dualcast 3.6/3.2
- Zap 2.0/1.5
- Strike+ 1.9/1.7
- Strike 0.4/0.7
- Defend -1.1/-1.1
- Self Repair -5.9/-5.8

It seems to think removing a strike+ is a better idea than removing a strike. Or maybe that Dualcast is a amazing card in my deck, and drawing Self Repair is like punching myself for 6, and the idiot that I am should just remove it and favor my Defends.

Weird.

this mod is helping me learn good synergies thankyou

i would use it but i use mods a lot and i want to use mods with this,

Is this mod working for people after the newest Slay the Spire update? This mod crashes the game due to an error caused by tensorflow for me.

I love seeing the Expected Damage during fights! Sometimes, I use this mod with Spire with Friends, and unfortunately, that mod places the players' name cards in a way that obscures Slay-I's expected damage. Is there any way to add a "vertical offset" setting that would allow me to change the positioning of the Slay-I text so I could see it when I use this mod?

Here's an example of the obscuring:
https://i.imgur.com/RjvfNdt.jpg

This is really cool

why does this crash

Please add an option to disable each elements. I want to view the expected dmg take but not the card stats

It told me to take a 2nd clash +4.8 96th percentile

It told me that clash is 95th percentile (really really good) I'm gonna take it and see what happens.

found a bug: the next 3 combats will have 1 hp choice will break the algorithm and show that all card reward choices are bad.

Seems like Concentrate is really underrated by the tool for some reason. Maybe its just a player thing it helped me quite alot, but damn I would never have thought that bouncing flask can be this impactful.

should add stats to relic when possible. Maybe explain what numbers in shop remove mean and event choices as well.

So what's the purpose of the mod, though? You say all this stuff about how it evaluates and everything but...is that it? Is it just a glorified number tracker, that just tells you a predicted outcome as you play? Or does it play for you? Or does it give you suggested actions to achieve the best possible outcome? Like, you don't ever state the actual PURPOSE of the mod, you just tell us what it evaluates. Is that ALL it does, is evaluate?

so.... what exactly does this do?

FYI, crashes on M1 Macs with no other workshop add-ons (except the base one)

Upon launching the game with this mod enabled, I can get into a run perfectly fine, loading in a character, etc. But as soon as I select an encounter, the game crashes and hard-locks the run.

how to toggle it?

Upgrading at the campfire sometimes crashes the game

neat idea, unfortunately it keeps crashing my game.

Keeps crashing my game

useful :D:8bitheart:

expected damga: 1
real damage: 40.
i feel like a dissapointment

Me, downloading this mod for the first time: I do as the crystal guides.
Me, after doing one run: The crystal is bad at guiding.

I mean, it got me most of the way through the second act, so I guess I can't complain. But I think it may overestimate defensive capabilities. It expected me to take 6 damage on Lagavulin. LAGAVULIN! Who takes 6 damage on Lagavulin?! Especially with a 17 card deck and no clear archetype.

it doesn't show the evaluations when offered cards off of dream catcher

@TerrenceM Thanks for reporting that bug. Just pushed a release that should fix that crash!

@Zu I would love to incorporate turn 1 draw into the model, but the data is not available to train on for that. I'm using run history files, which don't include card draw data

Had some crashes with the sensory stone event yesterday that were resolved after disabling Slay-I

This is an interesting case study in how the average spirelogs player seems to operate. I'm not sure whether I should find some of the numbers funny or concerning, or whether it's simply a coding limitation in the sense that the ai can only consider this many factors. Either way, I'd try to incorporate your turn 1 draw into the damage prediction. It's wildly off at times because of this.

Same as @L Ψ Kongroo on Reckless Charge but the difference is huge.
https://imgur.com/Kldn7U4
I had also a score of 3+ for buying Reckless Charge+ on Act 1, which is quite strange.

@kidcharlemagne Usually the predicted damage is pretty close (+/- 7HP). There may be some corner cases where you have cards in your deck that don't have a lot of data on them (eg. Panic Button). This can cause the model to make obviously inaccurate predictions

@L Ψ Kongroo The model treats upgraded cards as entirely different cards. Check out the medium article for more explanation as to why they are treated as entirely different cards!

https://steamcommunity.com/sharedfiles/filedetails/?id=2170515543

Why does the mod give a higher value for removing Strike+ over Strike?

Can anyone confirm the accuracy of the predicted damage? I was getting 1 damage on Gremlin leader and also The Champ. Either I reallllllly suck or that number is way the hell off.

@aldriedger here’s the link https://imgur.com/gallery/1ib9TSP

Unable to load jnitensorflow.dll due to my CPU doesn't support AVX.
Please update a option that using downgraded tensorflow(<1.5).

Update is live! Added scores in more screens (shop, library, and transform). Also the weighting has been updated to use a new weighted average that favours cards that do better in fights that you currently take a lot of damage in

@PrimeSonic I'll see what I can do about conveying more information about a cards upgrade value. I agree that it's an area where it would be interesting to know what the model thinks about it

Consider that you can see how many fires are remaining. Given the map information, you can estimate the 'number of upgrades still expected', and even when they are expected to hit.

Example: 2.5 more upgrades remaining in the act (from events and fires). Weakest provides est +1.2 hp according to slayAI.

Adding 'uppercut' would be a -1 (just considering the raw deck), while upgrading that would be a +3. So the upgrade over upgrading the other cards in the deck is an increase in value of +1.8. That leads me to believe the correct value (factoring in the increased value of upgrades in the future) is +0.8.

The constant to calculate/estimate is "what's the value of the remaining upgrades in the act?"
Ideally, consider skip: taking many cards necessarily means not upgrading everything. Value upgrades 'in the future'. Machine learning can help again?

I must also agree with @Karthas after trying out a run with The Defect and seeing so many recommendations to remove cards that I know are great when upgraded.

I think simply providing more information would be the key.
Show one line of numbers with the label "Remove" and another line of numbers with the label "Upgrade".
The same could be shown when adding non-upgraded cards, simply show what the value would be when upgraded.

When there's inevitably a conflict, the player can make sense of things with better long term planning than the numbers (currently) suggest.

@Karthas I've definitely thought about the behavior when a card is flagged as a good removal and a good upgrade. I like where you are going with the idea of incorporating the upgrade value into the score of the card and score of the removal. If you come up a good formula for how to do it, I would be interested!

@Boomer @Karthas I'm currently working on a model that should (hopefully) deal a little bit with the problems related to the model treating upgrades as different cards. It's called an Embedding layer, which essentially maps a card to vector. The model learns how to map a card to a vector but should theoretically map similar cards to similar vectors.

I feel like there's a fundamental problem with treating upgraded and un-upgraded cards as separate.

If you have a card that is a common sense "always upgrade this card" there will be very few runs/fights that make use of the un-upgraded variant, and any comparison would find that players who have that card in their deck (they didn't upgrade it), are taking more damage.

The reason this is a problem is that Eruption is both flagged as a highly desirable card to remove (around +3), and a highly desirable card to upgrade (also around +3).

I feel like a naive solution (one that doesn't require a major rework) is to offset the removal weight by the inverse of the upgrade weight. (Ideally there would also be some sort of adjustment for the evaluation of ADDING a card based on how good upgrading it would be)