Germaniac said:Now, evaluating at both defensive and offensive aspects is rather tricky. My idea is to have each of the 171 typings face off one another. During each matches, say between typing A/B against typing C/D (A the same as B and C the same as D, to account monotypes), I calculate which of A and B has the better effectiveness (x4, x2, x1, x0.5, x0.25, or x0) against typing C/D, let's call this effectiveness as max(A/B). Likewise I calculate max(C/D), then compare max(A/B) with max(C/D). If max(A/B)>max(C/D) then typing A/B wins the match, and if vice versa then typing C/D wins the match, otherwise it's a draw.
With number of wins minus number of losses as the basis for the ranking, the best typings are Flying/Steel (96 minus 35, also the highest number of wins), Flying/Ground, and Ground/Fire (both 91 minus 30). The lowest number of losses is Water/Dragon with 17.
I've been mulling over what measure to use as a tiebreaker. The easy way is to use number of wins, but I was wondering if there's a way to measure how big of a win each win is. The answer I arrived at is basically
how much health the winning side has at the end of the match.
Explanation
To understand it, we need to go back to how we defined a
match. It used to be just a comparison between the best effectiveness each side has against the other side. This time, imagine these:
- Each side is represented by a Pokemon.
- These Pokemon have 4 HP.
- These Pokemon have 1 move for each of its type, so mono-type will have 1 move and dual-type will have 2.
- Each move deals 1 HP multiplied by its type effectiveness against the defending Pokemon.
- HP and damage dealt can be non-integer.
- Every turn, each Pokemon will use the move that has best type effectiveness against the defending Pokemon.
- Every turn, attacks from both sides happen simultaneously.
- Keep doing this until one side faints.
- If both sides are immune to each other, assume that they faint simultaneously.
At the end of the match, we'll take the proportion (think percentage, but represented as 0 to 1 instead of 0% to 100%) of the winning side's remaining HP. This number is then added to both sides in the table, named as "Win Score" for the winning side and "Loss Score" for the losing side.
As an example, imagine a match between
Flying/Ground and
Ground/Fire:
- Turn 1: Flying/Ground uses Ground move, dealing 2 HP of damage (2 HP left). Ground/Fire uses Fire move, dealing 1 HP of damage (3 HP left).
- Turn 2: Flying/Ground uses Ground move, dealing 2 HP of damage (0 HP left). Ground/Fire uses Fire move, dealing 1 HP of damage (2 HP left).
- Ground/Fire has fainted, Flying/Ground has 2 HP left.
- 2/4 = 0.5 is added to Flying/Ground's "Win Score" and Ground/Fire's "Loss Score".
You may have noticed that the remaining HP proportion is entirely dependent on the best type effectiveness each side has against one another, which I denoted by max(A/B) and max(C/D). You can see the value of remaining HP proportion on the table below.
| Effectiveness | 0 | 0.25 | 0.5 | 1 | 2 | 4 |
| 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| 0.25 | 1 | 0 | 0.5 | 0.75 | 0.875 | 0.9375 |
| 0.5 | 1 | 0.5 | 0 | 0.5 | 0.75 | 0.875 |
| 1 | 1 | 0.75 | 0.5 | 0 | 0.5 | 0.75 |
| 2 | 1 | 0.875 | 0.75 | 0.5 | 0 | 0.5 |
| 4 | 1 | 0.9375 | 0.875 | 0.75 | 0.5 | 0 |
As you can see, the remaining HP proportion is actually equivalent to 1 - (max(C/D) / max(A/B)) where A/B is the winner of the match A/B vs C/D. This is what I use in my calculation, instead of actually simulating a match.
This method should give the same number of Win-Draw-Loss as our old method. But, now we also have a measurement of how big of a win each win is (Win Score) and how big of a loss each loss is (Loss Score).
With that covered, I recalculated the table. With number of wins minus number of losses (henceforth called "Result Difference") as the basis for the ranking and "Win Score" minus "Loss Score" (henceforth called "Score Difference") as the tiebreaker, our new sole #1 is
Flying/Steel. This is expected because both the other old tied-#1 has 4x weaknesses, amplifying their "Loss Score".
Interestingly, our old tied-#1
Flying/Ground and
Ground/Fire ties again here. They have the same number of Win-Draw-Loss, so the next fitting tiebreaker is probably either head-to-head or "Win Score" alone.
Flying/Ground wins on head-to-head while
Ground/Fire has the higher "Win Score".
Another interesting way to see the new table is what if we use "Score Difference" as the basis and "Result Difference" as the tiebreaker instead. In this ranking, the highly-regarded
Steel/Fairy rises from #7 to #4 and
Ground/Steel, which can be seen as the combination of arguably the best offensive and best defensive type, rises from #8 to #5. There are a lot of moving around in the table, but our top 3 remains the same. The main takeaway is this ranking highly rewards types with immunities and punishes types with x4 weaknesses.
The table can be checked
here.
------------------
There's one other thing I mulled over. My table seems to massively underrate mono-types. The highest ranked mono-type on the table is
Fairy at #98 with "Result Difference" as the basis, and
Steel at #100 with "Score Difference" as the basis.
This seems to stem from the fact that they only have 1 choice for their attack, meanwhile the dual-types can choose the better of their 2 attacks and just disregard the other attack for the whole match.
The solution I came up for this is to take the effectiveness from both types in dual-types and add them together. For mono-types I just double the effectiveness, this is equivalent to treating them as dual-types, but has the same first and second type.
Explanation #2
If you read the Explanation spoiler above, this time think of the Pokemon starting with 6120 HP and using 2 moves, 1 of each type, every turn. Yes, this means there are 4 moves used in each turn and they all happen simultaneously.
Why 6120? It's the LCM of (0.25, 0.5, 0.75, 1, 1.25, 1.5, 2, 2.25, 2.5, 3, 4, 4.25, 4.5, 5, 6, 8), all the possible damage dealt in a single turn, except 0. This makes sure that each match ends with one side dealing damage exactly equal to the remaining HP of the opponent.
Like the previous method, I use a much simpler formula for the calculation, but this time using sum instead of max.
You'd think that this is even more impractical because, why would a Pokemon ever use its weaker attack? But, I find the results interesting.
With "Result Difference" as the basis and "Score Difference" as the tiebreaker, our #1 is...
Flying/Steel again. But, this time it wins by a landslide, with a "Result Difference" of 74 compared to 65 of the second place,
Steel/Fairy. Rounding out the top 5 are
Steel/Water (62),
Flying/Ground (61), and
Steel/Dragon (60).
Swapping the basis and the tiebreaker for the ranking gives a top 5 of
Flying/Steel,
Steel/Fairy,
Ghost/Steel,
Steel/Water, and
Ground/Steel.
This new method does solve the problem of mono-types being underrated, with the highest mono-type being
Ghost at #17 with "Result Difference" as the basis, and
Steel at #23 with "Score Difference" as the basis.
The table can be checked
here.
------------------
One of the biggest talking points of ranking the 171 typings is how much weight should one give to the offensive side of the typing. After all, a better type effectiveness always results in higher damage than having STAB. So, a Pokemon can just have the best defensive typing and use the best offensive types for its attacks.
How do we find the best defensive typing? I've seen a few people rating each typing by averaging the type effectiveness it has defensively against the 18 offensive types, and the sorting the results from lowest to highest.
That method does give good result. But, I want to take one step further, I want to take into account that the better offensive types will see higher usage.
Explanation #3
How do we do it? We'll give a weight to each offensive type, representing how good they're as an offensive type. How do we calculate this weight? We use the same idea we used defensively, we average the type effectiveness each offensive type has against all the 171 possible type combination, let's call this "Offensive Score".
Now, we can get a new defensive typing by averaging the type effectiveness, but this time we weight (multiply) each offensive type by the "Offensive Score" they have.
But, we won't stop here. Since the better defensive type will also see higher usage, we'll take the defensive type effectiveness average to update our "Offensive Score". But, note that the average is reversed, i.e. the higher average a typing has, the worse typing it is. To account for this, I'll use the reciprocal of the number, and we'll call this "Defensive Score".
Similar to how we updated our "Defensive Score", we'll update our "Offensive Score" by averaging type effectiveness, weighted by each typing "Defensive Score. After this, we can update our "Defensive Score" again.
So, in short, we use "Defensive Score" (for 171 typings) to calculate "Offensive Score" (for 18 types), then use that "Offensive Score" to update the "Defensive Score" which we can use to update the "Offensive Score" again. We can repeat this as many times as we want. But, I found that after doing 25 iterations the ranking has converged.
What's the best defensive typing we get by this method? Surprise, it's
Flying/Steel again. Our second place is
Steel/Fairy. People would've expected it to get the first place, and indeed it would've been first if we only did 1 iteration (i.e. all "Offensive Score" are 1). I guess changing Ground weakness into immunity gives a big swing to the score. Rounding out the top 5 are
Steel/Water,
Steel/Dragon, and
Ghost/Steel.
You can check the "Defensive Score" numbers
here (the final score is on the rightmost column, named Iteration 25).
Note that the numbers are oscillating in period 2. That's why I said the
ranking converged, instead of the numbers. I guess this happened because I was taking reciprocals. But, don't worry, the ranking is still the same even if you sort by Iteration 24.
This method also gave me the ranking for offensive types, in order they're Ground, Rock, Fire, Ice, Water, Fairy, Flying, Fighting, Steel, Dark, Ghost, Psychic, Electric, Dragon, Grass, Bug, Poison, Normal. You can see the "Offensive Score" numbers
here.
EDIT: By treating Offensive Score as probability (constraining the sum to 1) I managed to make the numbers converge. The new Defensive Score sheet is
here and new Offensive Score sheet is
here. The ranking doesn't change.