Zelda Data

[schrum2]

Scan The Legend of Zelda's data. When I did this with the GAN before, I focused on individual rooms. That can certainly be done here too, and should perhaps be tried as a starting point, but I wonder if we could get away with scanning the dungeons, and generating whole dungeons. That would be an extra challenge. Also, it's not a platformer, so there are extra issues to worry about in testing.

[schrum2]

Ok, I have a new idea for how to handle the Zelda data. A dataset of individual rooms is too boring, and doesn't give us a great way to build up dungeons. Instead, make a dataset of 2 by 2 clusters of up to 4 rooms (some may be absent). Use the create_level_json_data.py file to do this, but you will need to modify/generalize it in a way to work on Zelda but still work on Mario and Lode Runner #39 as well. In fact, this might conflict with some work that @williamsr03 did in this file, so work with him on it.

[besshagan]

[] I talked to Reid about his work on LoadRunner (#39 ) and I am currently doing some preliminary research to familiarize myself with the game and the data.

[] DATA MAPPING (w/ notes):

F = FLOOR
B = BLOCK
(movable blocks are also labelled B)

M = MONSTER
P = ELEMENT (LAVA, WATER)
O = ELEMENT + FLOOR (LAVA/BLOCK, WATER/BLOCK)
(walkable floor in first level beginning room)

I = ELEMENT + BLOCK
(I appears for the first time in tloz4_1.txt)

D = DOOR
(May be useful for understanding the size of rooms)

S = STAIR
(rare feature in rooms)

W = WALL

• = VOID

[] Familiarizing myself with game and data

• Looked at gameplay for first level dungeon on YouTube: https://youtu.be/6g2vk8Gudqs?si=FWI5AIwjOZ_Cl_5S

[] Adjusted Original pngs to match Processed text file representations
[] Manually removed M (monsters) and replaced these with B (blocks) in the first room of the level 1 dungeon
[] Pushed changes to main.

[besshagan]

[] I found that the first section of each dungeon uses a similar pattern of text symbols across all levels, so there were M (monsters) in places that there should be.
[] Manually removed M (monsters) and replaced these with B (blocks) in the text files for all dungeons.
[] I think I understand the data a bit better now. Below are some notes I wanted to keep for later:

width = 11
length = 16

[besshagan]

[] Worked with GPT 4.1 to find a way to scan through Zelda data
[] The code runs, but the output is incompatible with the ascii data browser
[] Created a new JSON from the Zelda READ ME to map things correctly
[] Attempted to add empty captions using Reed's script: LR_create_ascii_captions.py to assign empty captions (saved to file tloz_Levels-and-captions.json.)
[] Added print statements to create_level_json_data and ran the code again to try to get a better idea about what might be going wrong. (See sample of printed output below)

python create_level_json_data.py --tileset "C:\Users\haganb\Documents\GitHub\TheVGLC\The Legend of Zelda\zelda_tileset.json" --levels "C:\Users\haganb\Documents\GitHub\TheVGLC\The Legend of Zelda\Processed" --output "tloz_Levels.json" --scan_mode room_cluster

Level size: 10x9
Room size: 16x11
Dungeon grid: 0 rows x 0 cols
Padded grid: 1 rows x 1 cols
Padded level height: 16, width: 11
Extracted 0 clusters
Level size: 96x66
Room size: 16x11
Dungeon grid: 6 rows x 6 cols
Padded grid: 7 rows x 7 cols
Padded level height: 112, width: 77
Extracted 36 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room
Level size: 32x88
Room size: 16x11
Dungeon grid: 2 rows x 8 cols
Padded grid: 3 rows x 9 cols
Padded level height: 48, width: 99
Extracted 16 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room
Level size: 64x88
Room size: 16x11
Dungeon grid: 4 rows x 8 cols
Padded grid: 5 rows x 9 cols
Padded level height: 80, width: 99
Extracted 32 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room
Level size: 48x88
Room size: 16x11
Dungeon grid: 3 rows x 8 cols
Padded grid: 4 rows x 9 cols
Padded level height: 64, width: 99
Extracted 24 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room
Level size: 80x66
Room size: 16x11
Dungeon grid: 5 rows x 6 cols
Padded grid: 6 rows x 7 cols
Padded level height: 96, width: 77
Extracted 30 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room
Level size: 64x77
Room size: 16x11
Dungeon grid: 4 rows x 7 cols
Padded grid: 5 rows x 8 cols
Padded level height: 80, width: 88
Extracted 28 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room

[besshagan]

[] These outputs appear to be in line with the file order in Processed, beginning with the READ ME, where no 0 clusters are extracted.
[] However, when I count how many clusters I expect to be extracted, the numbers don't make a lot of sense to me when examining them against the levels. So, I think the padding might be affecting the way this code is running.
[] This may or may not be part of the hashing problems that won't let me use the ASCII browser

[besshagan]

[] I left the following notes in a note pad for the next time I work on this issue:

ZELDA NOTES

Loading level file: C:\Users\haganb\Documents\GitHub\TheVGLC\The Legend of Zelda\Processed\tloz1_1.txt

Level size: 96x66
Room size: 16x11
Dungeon grid: 6 rows x 6 cols
Padded grid: 7 rows x 7 cols
Padded level height: 112, width: 77
Extracted 36 clusters
Sample cluster shape: 4 rooms, 16 rows per room, 11 cols per room

---

22 cluster without padding

is padding being added to all sides?
if so, then window may not contain whole rooms....

if 1 extra row and column are added entirely, then there are too many clusters....