Yacht Dice: Fix logic (again) so that score doesn't drop when receivi…

…ng item (#4044)

* Add the yacht dice (from other git) world to the yacht dice fork

* Update .gitignore

* Removed zillion because it doesn't work

* Update .gitignore

* added zillion again...

* Now you can have 0 extra fragments

* Added alt categories, also options

* Added item categories

* Extra categories are now working! 🐶

* changed options and added exceptions

* Testing if I change the generate.py

* Revert "Testing if I change the generate.py"

This reverts commit 7c2b3df.

* ignore gitignore

* Delete .gitignore

* Update .gitignore

* Update .gitignore

* Update logic, added multiplicative categories

* Changed difficulties

* Update offline mode so that it works again

* Adjusted difficulty

* New version of the apworld, with 1000 as final score, always

Will still need to check difficulty and weights of adding items.
Website is not ready yet, so this version is not usable yet :)

* Changed yaml and small bug fixes

Fix when goal and max are same
Options: changed chance to weight

* no changes, just whitespaces

* changed how logic works

Now you put an array of mults and the cpu gets a couple of tries

* Changed logic, tweaked a bit too

* Preparation for 2.0

* logic tweak

* Logic for alt categories properly now

* Update setup_en.md

* Update en_YachtDice.md

* Improve performance of add_distributions

* Formatting style

* restore gitignore to APMW

* Tweaked generation parameters and methods

* Version 2.0.3

manual input option
max score in logic always 2.0.3
faster gen

* Comments and editing

* Renamed setup guide

* Improved create_items code

* init of locations: remove self.event line

* Moved setting early items to generate_early

* Add my name to CODEOWNERS

* Added Yacht Dice to the readme in list of games

* Improve performance of Yacht Dice

* newline

* Improve typing

* This is actually just slower lol

* Update worlds/yachtdice/Items.py

Co-authored-by: Exempt-Medic <[email protected]>

* Apply suggestions from code review

Co-authored-by: Exempt-Medic <[email protected]>

* Update Options.py

* Styling

* finished text whichstory option

* removed roll and rollfragments; not used

* import; worlds not world :)

* Option groups!

* ruff styling, fix

* ruff format styling!

* styling and capitalization of options

* small comment

* Cleaned up the "state_is_a_list" a little bit

* RUFF 🐶

* Changed filling the itempool for efficiency

Now, we start with 17 extra items in the item pool, it's quite likely you need at least 17 items (~80%?).
And then afterwards, we delete items if we overshoot the target of 1000, and add items if we haven't reached an achievable score of 1000 yet. Also, no need to recompute the entire logic when adding points.

* 🐶

* Removed plando "fix"

* Changed indent of score multiplier

* faster location function

* Comments to docstrings

* fixed making location closest to goal_score be goal_score

* options format

* iterate keys and values of a dict together

* small optimization ListState

* faster collection of categories

* return arguments instead of making a list (will 🐶 later)

* Instead of turning it into a tuple, you can just make a tuple literal

* remove .keys()

* change .random and used enumerate

* some readability improvements

* Remove location "0", we don't use that one

* Remove lookup_id_to_name entirely

I for sure don't use it, and as far as I know it's not one of the mandatory functions for AP, these are item_name_to_id and location_name_to_id.

* .append instead of += for single items, percentile function changed

Also an extra comment for location ids.

* remove ) too many

* Removed sorted from category list

* Hash categories (which makes it slower :( )

Maybe I messed up or misunderstood...
I'll revert this right away since it is 2x slower, probably because of sorted instead of sort?

* Revert "Hash categories (which makes it slower :( )"

This reverts commit 34f2c1a.

* temporary push: 40% faster generation test

Small changes in logic make the generation 40% faster.
I'll have to think about how big the changes are. I suspect they are rather limited.
If this is the way to go, I'll remove the temp file and redo the YachtWeights file, I'll remove the functions there and just put the new weights here.

* Add Points item category

* Reverse changes of bad idea :)

* ruff 🐶

* Use numpy and pmf function to speed up gen

Numpy has a built-in way to sum probability mass functions (pmf).
This shaves of 60% of the generation time :D

* Revert "Use numpy and pmf function to speed up gen"

This reverts commit 9290191.

* Step inbetween to change the weights

* Changed the weights to make it faster

135 -> 81 seconds on 100 random yamls

* Adjusted max_dist, split dice_simulation function

* Removed nonlocal and pass arguments instead

* Change "weight-lists" to Dict[str, float]

* Removed the return from ini_locations.

Also added explanations to cat_weights

* Choice options; dont'use .value (will ruff later)

* Only put important options in slotdata

* 🐶

* Add Dict import

* Split the cache per player, limit size to 400.

* 🐶

* added , because of style

* Update apworld version to 2.0.6

2.0.5 is the apworld I released on github to be tested
I never separately released 2.0.4.

* Multiple smaller code improvements

- changed names in YachtWeights so we don't need to translate them in Rules anymore
- we now remember which categories are present in the game, and also put this in slotdata. This we do because only one of two categories is present in a game. If for some reason both are present (plando/getitem/startinventory), we now know which category to ignore
-

* 🐶 ruff

* Mostly minimize_extra_items improvements

- Change logic, generation is now even faster (0.6s per default yaml).
- Made the option 'minimize_extra_items' do a lot more, hopefully this makes the impact of Yacht Dice a little bit less, if you want that. Here's what is also does now:
 - you start with 2 dice and 2 rolls
 - there will be less locations/items at the start of you game

* ruff 🐶

* Removed printing options

* Reworded some option descriptions

* Yacht Dice: setup: change release-link to latest

On the installation page, link to the latest release, instead of the page with all releases

* Several fixes and changes

-change apworld version
-Removed the extra roll (this was not intended)
-change extra_points_added to a mutable list to that it actually does something
-removed variables multipliers_added and items_added
-Rules, don't order by quantity, just by mean_score
-Changed the weights in general to make it faster

* 🐶

* Revert setup to what it was (latest, without S)

* remove temp weights file, shouldn't be here

* Made sure that there is not too many step score multipliers.

Too many step score multipliers lead to gen fails too, probably because you need many categories for them to actually help a lot. So it's hard to use them at the start of the game.

* add filler item name

* Textual fixes and changes

* Remove Victory item and use event instead.

* Revert "Remove Victory item and use event instead."

This reverts commit c2f7d67.

* Revert "Textual fixes and changes"

This reverts commit e9432f9.

* Remove Victory item and make it an event instead

* Yacht Dice logic fix, no decreasing score when obtain item

take 2

* Logic fix: Revert max_tries and mults, change ordering

* Remove spaces :^)

* Updated weights that are stochastically ordered by dice/roll

In the trimming of the weights, sometimes it having 4 rolls would be better than having 5 rolls.
I did a check that this does not happen for any dice increment or roll increment

* Swap for-loops to increase performance

This method is faster if the first for-loop contains fewer items.
Since the function is called with, typically, `dist2` having less items, let's loop over `dist2` first. This makes the entire program 10% faster.

* Remove options with 0 chance from list

---------

Co-authored-by: NewSoupVi <[email protected]>
Co-authored-by: Exempt-Medic <[email protected]>

worlds/yachtdice/Rules.py

@@ -101,14 +101,15 @@ def dice_simulation_strings(categories, num_dice, num_rolls, fixed_mult, step_mu
         return yachtdice_cache[player][tup]
 
     # sort categories because for the step multiplier, you will want low-scoring categories first
-    categories.sort(key=lambda category: category.mean_score(num_dice, num_rolls))
+    # to avoid errors with order changing when obtaining rolls, we order assuming 4 rolls
+    categories.sort(key=lambda category: category.mean_score(num_dice, 4))
 
     # function to add two discrete distribution.
     # defaultdict is a dict where you don't need to check if an id is present, you can just use += (lot faster)
     def add_distributions(dist1, dist2):
         combined_dist = defaultdict(float)
-        for val1, prob1 in dist1.items():
-            for val2, prob2 in dist2.items():
+        for val2, prob2 in dist2.items():
+            for val1, prob1 in dist1.items():
                 combined_dist[val1 + val2] += prob1 * prob2
         return dict(combined_dist)