This fit, fits in game but not in pyfa.

[Neugeniko]

This uses 87.52/87.5 pg(doesn't fit) in pyfa but in game appears to use 87.48/87.5 pg(fits). It appears each modules pg is rounded to 2 decimal places in game before being added up to achieve this.

[Thrasher, Gank ART - Neug Kin]

Gyrostabilizer II Gyrostabilizer II

1MN Y-S8 Compact Afterburner Small Shield Booster II Upgraded Kinetic Deflection Amplifier I

280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S 280mm Howitzer Artillery II, Republic Fleet Titanium Sabot S [Empty High slot]

Small Core Defense Capacitor Safeguard I Small Projectile Burst Aerator II Small Anti-Thermal Screen Reinforcer I

Republic Fleet EMP S x3000

[DarkFenX]

I thought I've implemented similar logic in cbd1a34c685d55b01646e33ac541d020195b252a, but probably it doesn't apply in all cases. I will take a look at it later.

[DarkFenX]

Not defending my code (yet), but for guns 11×0.9×1.05 = 10.395. Strictly speaking, according to CCP's logic it should be rounded up to 10.4 pg. But this one is quite peculiar value: depending on accumulated float errors (which depends on order of operations) it might go either way, roughly speaking 10.394444444444444444449 turns into 10.39, and all values starting from 10.395 and higher should result in 10.4.

Are you confident that this fit consistently fits in EVE? Maybe there're some occasions when it doesn't? Can you try to unfit some guns and fit again, or try another ship and see if result is consistently the same?

ps Not defending my code because it seems to take unrounded value - 9+3+0.75+1+1+10.395000000000001×7 which is rounded up to 87.2 in the end. But if I implement this code properly we will end up with 9+3+0.75+1+1+10.4×7 which is 87.55, which is even worse.

[DarkFenX]

I fixed the change i linked earlier in 7b8d9f8dbe0a1064ceb91bc9a84d2587154a5e71, now attributes are rounded properly (and yes i checked that this time). Powergrid consumption of linked fit is 87.55, as I expected.

Fixing it properly needs some serious research on how EVE handles operations and why EVE accumulates float errors this way, and pyfa accumulates the other way around.

I think we cannot 100% reliably know how EVE works with all these numbers. However, we can write specific logic so that if round(n, 9) == x.xx5, then number is rounded down. The question is - if we do that, will there be any fits which fit in pyfa but do not fit in game?

Most research should be concentrated on finding such combinations of modules/rigs/implants when item's cpu/pg equals x.xx5 then.

[DarkFenX]

I will put here examples I find:

• any turret with odd powergrid consumption (1, 3, 5, ...) on a ship without bonus/malus to pg, with 1 weapon rig
• ships with cpu/pg output of 2+4*n, cpu/pg skill V and 1%/3%/5% cpu/pg implant
• some ships with pg skill 5, t1 ACR and 1% pg implant (Adrestia 1783.155, Succubus 61.105)
• ships with odd cpu/pg output, cpu/pg skill V and 2%/6% cpu/pg implant
• hybrid/energy turret cpu consumption, if its base consumption is 2 + 4*n (2, 6, 10, ...), with weapon upgrades V and t2 algid rig
• ships which have pg output of 8 + 16*n, with pg skill V, t2 acr rig, 1% pg implant
• any turret with base cpu consumption of 2+4*n, with weapon upgrades V and 1%/3%/5% wu- implant

[DarkFenX]

Results of some tests.

what	decimal formula	decimal result	eve result	python result
280mm arty pg	11×0.9×1.05	10.395	d10.39	d10.39
Polarized heavy pulse pg with 1 weapon rig	189×0.9×1.05	178.605	d178.6	d178.6
DG 250mm rail pg with 1 weapon rig	179×0.9×1.05	169.155	u169.16	u169.16
Retribution PG with 1% imp	62×1.25×1.01	78.275	u78.28	u78.28
Retribution PG with 3% imp	62×1.25×1.03	79.825	u79.83	u79.83
Retribution PG with 5% imp	62×1.25×1.05	81.375	u81.38	u81.38
Cruor CPU with 1% imp	150×1.25×1.01	189.375	u189.38	u189.38
Cruor CPU with 3% imp	150×1.25×1.03	193.125	u193.13	d193.12
Cruor CPU with 5% imp	150×1.25×1.05	196.875	u196.88	u196.88
Succubus PG with t1 acr with 1% imp	44×1.25×1.1×1.01	61.105	u61.11	d61.1
Celestis PG with 2% imp	575×1.25×1.02	733.125	d733.12	d733.12
Celestis PG with 6% imp	575×1.25×1.06	761.875	u761.88	u761.88
Celestis CPU with 2% imp	375×1.25×1.02	478.125	d478.12	d478.12
Celestis CPU with 6% imp	375×1.25×1.06	496.875	u496.88	u496.88

Python result is round(n, 2) of decimal result, d means rounding down, u means rounding up. It is clear that in both areas final result is dictated by float error. We might try following approaches:

1) Try to reverse-engineer how EVE processes calculations and implement the same way. It's huge :effort:, alot of testing and many restrictions applied to how pyfa handles calculations 2) Implement decimal-alike logic. I.e. in all cases mentioned above we will round up. It will not fix original issue though. 3) Do nothing and leave as-is. EVE has its own float errors, we have our own.

[DarkFenX]

So, in case of arty. List of involved items:

• 280mm Howitzer Artillery II
  • attribute "power" has base value of 11.0
• Small Projectile Burst Aerator II
  • attribute "drawback" has base value of 10.0
  • has effect "drawbackPowerNeedProjectiles" (applies PostPercent modification using value of "drawback" to items from "Projectile Weapon" group, attribute "power")
• Projectile Weapon Rigging
  • attribute "rigDrawbackBonus" has base value of -10.0
  • has effect "rigDrawbackSkillEffect" (applies PreMul modification using value of "skillLevel" to self, attribute "rigDrawbackBonus")
  • has effect "rigDrawbackReductionProjectile" (applies PostPercent modification using value of "rigDrawbackBonus" to items from "Rig Projectile Weapon" group, attribute "drawback")
• Advanced Weapon Upgrades
  • attribute "powerNeedBonus" has base value of -2.0
  • has effect "sensorUpgradesSkillBoostPowerNeedBonus" (applies PreMul modification using value of "skillLevel" to self, attribute "powerNeedBonus")
  • has effect "skillAdvancedWeaponUpgradesPowerNeedBonus" (applies PostPercent modification using value of "powerNeedBonus" to items which have "Gunnery" as their skill requirement, attribute "power")

    
    pg = 11.0
    rig_drawback = 10.0
    skill_rig_drawback_bonus = -10.0
    awu_bonus = -2.0

rig_skill_effect = skill_rig_drawback_bonus 5.0 adjusted_drawback = rig_drawback (1 + rig_skill_effect / 100) awu_skill_effect = awu_bonus * 5.0

result = pg (1 + awu_skill_effect / 100.0) (1 + adjusted_drawback / 100)

print(result)

results in `10.395000000000001`

No idea how to change it so that result matches ingame.

[DarkFenX]

Some extra info:

• http://lukewrites.com/posts/rounding-decimals-in-python-or-why-doesnt-5-round-to-1/
• https://en.wikipedia.org/wiki/Rounding#Round_half_to_even

[DarkFenX]

For logging: a while ago I've filled EVE issue EBR-175339 which concerns inconsistent attribute rounding. Heard no feedback on it so far.