Revision as of 14:07, 31 May 2019

https://twitter.com/EmojiTetra/with_replies

Some analysis

• Cool graphs: http://pub.thorx.net/emojitetra/
• EmojiTetra/DeepStatistics

Some plans for future analysis

Board health

Ideas to measure board health

• simple board fill - Count of positions on the board that are used. This is extremely simple and is already in the analysis
• modified board fill - count columns that have holes in them, and use that as modifying factor to the simple board fill. The height of those columns (or of highest?) should be a factor too
  • discussion in this thread: https://twitter.com/HeptaSean/status/1063093435719213056
• board age - for each used position, track how long (age in pieces spawned) it has been on the board. Health is then the sum of ages. Lower = healthier
  • A low board will have a higher turnover of pieces and better health. A high board will have low rows that are stuck for a long time - as they age the health will decrease, indicating we are high and possibly stuck there...
  • however, it's still not an objective absolute, as two identical boards may register different health
  • discussion in this thread: https://twitter.com/Caber_nl/status/1065174652060909570
• weighted board fill - count occupied positions in columns. The unhealth of a piece is valued according to how difficult it is to get to. Basically, the more buried a piece is, the higher it scores. Buried holes need to be handled in some form (they are bad), and deep open wells need to be handled (they are good)
  • possible implementation: every occupied position is a point. Those that are on an equal row or higher than the lowest accessible open spot have no further weighting, however each occupied position below that row gets an additional +1 for each row further down. Each unoccupied position (hole!) gets an additional +1.
  • Should it handle horizontal holes (ie don't count quite as strongly because once uncovered, they'll be easier to fill and move on?)

Sociopolitical analysis

The game can be examined as a microcosm of human democratic/communication behaviour!

I wrote some notes about this at one point. There have been many tweets too.

• For instance, this thread: https://twitter.com/alexjc/status/1063861229536665601

Rhyming game tips

(refer also to My Terminology below)

• "less than four? build up more"
• "make a column only if it's to the bottom"
  • on the side build a column, but only if it's to the bottom
  • note: uses non-standard use of "column". Should find a tip to call it a well
• "more than five? clear some lines to stay alive"
  • I like it, except the meter in the first and second clauses is mismatched. “If you’re stuck with more than five, clear some lines to stay alive.”
    • https://twitter.com/JoeSondow/status/1031215485596004352

• "1+1<2" - this is great for sentiment and brevity, but the only people it would be clear to are the power players who know it anyway!

Game Table

Note: my timezone is UTC+10 and my dates reflect that. I hope to direct link to tweets sufficient that they can be used for exact time lookups if needed

• More details of games (list of pieces) are at EmojiTetra/Details

Board styles / known feature changes

1. Start of Game 1 from the 16 April 2018: score in blocks, then unicode SOON, then linedrawing to illustrate NEXT piece
2. During Game 1 on the 19 April: unicode SOON replaced with unicode F-FWD
3. During Game 5 on the 22 May: Literal "Next", then braille indicator of next piece. Then literal "Score" in and score in plain text
4. During Game 6 on 5 June: "Next" and score moved to right of board. Next illustrated with emoji same as board
5. During Game 18 on 10 Feb: "The bot will now tweet a rendition of the game grid with filled rows marked as disappearing, with the score benefit shown." - https://twitter.com/emojitetra/status/1094340439849615360

Notable events

Empty Board Bonus

Most are with doubles scoring 1250. Exceptions noted

• Game 6 with a T: https://twitter.com/EmojiTetra/status/1007392918284337153
• Game 8 with a T: https://twitter.com/EmojiTetra/status/1014615507675201536
• Game 15 with an L: https://twitter.com/EmojiTetra/status/1060517954972921858
• Game 15 with a T: https://twitter.com/EmojiTetra/status/1070141366217334786
• Game 15 with a J: https://twitter.com/EmojiTetra/status/1070257128940523522
• Game 17 with an L: https://twitter.com/EmojiTetra/status/1091275625296388096
• Game 19 with an I: https://twitter.com/EmojiTetra/status/1097119131365650432
  • (single - 1100 points)
• Game 25 with a J: https://twitter.com/EmojiTetra/status/1131878166207684609
  • (triple - 1525 points)

Awesome moves

For those times we manage to fill in overhangs, T-spins and I-spins

• • TODO

Longest activity on the same piece

(todo: backfill this with any other triple-spins)

Great Escapes

When The End Is Nigh - and then it wasn't. This is the list of times we've filled the board to 45 or more static pieces, and subsequently recovered to under 30. "Saved to" is measured as the board fill after clearing lines, to the point where two pieces in a row begin to stack again.

Other

This magnificent I handling in Game 23 shows how skillful the hivemind can be - when the non-obvious move is sufficiently spruiked into awareness. https://twitter.com/EmojiTetra/status/1123573444400914434

Scoring rate

Scoring rate vs total

Real world analysis

There is reasonable correlation with better scoring rates, and better scores. Indeed, as of writing (G24 just begun), there are 11 games with a rate above 1100/day - these have a min score of 11925 (avg: 32580), whilst 12 games below that rate have a maximum at merely 13850 (avg: 6373)

Within reason, this seems logical. Smart play is required for long games, but also improves on scoring rate by seeing non-obvious combinations of the current+next piece to score doubles and triples, where more naive play (as the hive often does) will seek singles only. Conversely, simple "just get singles" non-aggressive play will often result in situations where, thanks to a run of awkward pieces, the board gets messy and the hive simply hasn't learnt and practiced the skills to navigate out.

Hypothetical maximums

Hypothetically, the maximum scoring rate with ideal pieces is a board-emptying tetra. 2000 points every 7th piece. I believe the fastest this could be done is with 6 "O" and then one "I" bar. This would be nearly 7579 points per day - or over 6.3x our normal scoring rate of approx 1200points/day

At our normal randomised piece rate of placement, 2000/7pieces would still be approx 4x our normal scoring rate. But that is also impossible in practice, since it relies on us being able to build a 6wide 4high tower every 6 pieces, and then have an I to score the emptying tetra every 7th piece. In practice of course, we only get an I as the 7th piece 1/7th of the time. We also cannot get an empty board every 7th piece, even if we had prescient knowledge of the future. However, I think it should be possible to get approx 2x our native scoring rate from smart piece placement alone. How? By targetting the empty board bonus via full clears as the default tactic. This should be safe as it's a tactic which inherently means we keep the board low, we have a regular "seventh piece" cycle to play to, and if we build clean, then we should get plenty of doubles and triples and occasional tetras the rest of the time.

Safer AND faster scoring rate. What's not to love?

Without a way to play solo and test tactics, this is theoretical only. (A javascript implementation to allow faster one-person play would be great as a practice/education tool. Could even feed it the official game piece order for direct comparisons with the hive?)

My Terminology

(note: this is what I endeavour to use, and while I encourage others to use, it is in no way formal or endorsed. Most of it is what is commonly used already)

• board: play area (7x11 squares)
• column: a vertical sequence on the board from top to bottom (11 spots)
• line/row: a horizontal sequence on the board from left to right (7 spots)
• clear - to clear some rows
• single/double/triple: to clear 1/2/3 line(s)
• tetra/quad/tetris: to clear 4 lines
• collective/hive/electorate/crowd: everyone who votes
• hole (1): an unfilled space that is surrounded
• overhang/half-hole: an unfilled spare that is covered at top, but open to the side
  • some overhangs can be underfilled by a side slide. But not all
• clean - board without holes. as in "the board is clean" and "this is a clean drop"
  • question: is a half-hole a clean board?
• dirty - opposite of clean: a board with a hole. "the board is dirty". "this is a dirty drop"
• well/hole (2): a space perfectly suited for a single piece. Usually I-hole, sometimes J or L-hole.
• silhouette/skyline: the shape of the top surface
• empty board: nothing on the board. a perfectly flat skyline, as per the start of a game.
  • I prefer 'empty' to avoid possible ambiguity with clear or clean, however this is where my usage diverges most from the electorate who use "full clear".
• fossilised - to refer to how long a row has remained unchanged

Externals

Credits

• Game created by https://twitter.com/JoeSondow
  • Tweet with links to his other accounts too: https://twitter.com/JoeSondow/status/971561164994895872
• EmojiTetra game code on gitlab: https://gitlab.com/JoeSondow/emojitetra

Other/misc

• Animation view of game history: https://emojitetra-viewer.glitch.me
• A different set of stats/charts of game history, including poll analysis: https://emojitetra-viewer.glitch.me/stats