Saturday 16 September 2017

大逆轉裁判2 心得(有雷)

(完全劇透,慎入)

音樂

逆轉系列本身的音樂水準毋庸置疑。本集沿用一代風格,每一集都分到適當氣氛的音樂,老玩家期待舊有固定BGM的重製(Objection/緊張情景)也都在。當然對我這種不喜歡開音樂玩遊戲的人來說似乎沒有資格過於詳細地逐首評論。

畫風效果

模型比一代做得更多更仔細,人物也帶有一般濃厚的英倫風(不論是外型還是台詞),角色群拿到大量的動畫可以說是加分位--主角群有眼泛淚光的愛麗絲醬和踢踏舞悠仁,配角蠟像館夫人(好萌!)各種pose和夏目漱石的新連技也令人印象深刻。
問題是劇本本身對配角的刻畫不夠深入--模型只是刻畫一個角色的基本,缺乏故事本身的襯托的話也沒法讓玩家深深記住。另外前三章犯人暴走也不夠明顯也令人失望,撇除這兩點的話大逆轉2在美術方面是在合格線以上的。

逐集短評 

I
人啊,活在世上就會有意外發生。一個可以做出第五章級別案件的人這樣就被幹掉了。
作為教學關其實略顯違和。對於沒有逆轉經驗的玩家來說要在第一章就大量運用到法醫學常識其實不太容易(儘管有諸多提示),而對老玩家來說案件本身又過於簡單(沒有搜證就沒有展開的機會),反正去到一半發現了一個沒聽過的證人出來。那很好,有罪的人就決定是你了!用毒這點在一代初章也用過了,再用一次其實有點悶,不過手法本身倒是頗有趣。
以英國為舞台的遊戲出現了大量以英國角色看日本人(尤其是死神)的台詞十分合理,但相反日本角色在日常層面對英國的刻畫其實不多(包含一代),可惜這裡實在沒有太多機會讓巧舟發揮。
然後亞內你維新不快點明治就要變大正了……

II
以傳統逆轉角度來說做得最好的一章。事件本身基本上不涉及主角群,這樣配角就能分到大量的發揮空間;大量零散的提示看似凌亂,實則只是玩家處於不同時空難以理解;透過與犯人的長篇對決中將真相逐層剝開,最後逆轉出的事實則讓人拍案叫絕。
大家可以回想一下歷代逆轉裁判大多都是主動襲擊被害者(這也是理所當然的),佈局殺人最近大概就是逆轉6-2和4-4(霧人)了。從一代四章「必然的偶然」開始到本章兩起「偶然的必然」結束,這案件單純以推理的角度來說應該是最好的一章。
所以,羅密歐和侏麗葉那個更強呢?那一定是高崎美咲演的羅密歐--

III
萬博會。水晶宮(其實以現代的角度去看當年的水晶宮不太「水晶」)。變魔術。
案件本身其實只是變魔術。堅信瞬移是不可能的玩家幾乎可以一開始就做出這個結論。
既然是魔術,那只要調查裝置本身事故就不難解拆,本來是應該這樣的。巧舟不可能讓第三章如此容易,解拆裝置面前有座大山叫科學技術保護條例--這種圍繞與殺人無關的條例的討論可是算是冗長的法庭戰中的清流,同樣的例子還有逆轉6-5前段對於始祖寶玉所有權的爭論。當然遊戲討論不可能利用條文/契約本身的技術細節(這樣對一般玩家來說太難了),所以兩件案件的結論都由殺人案本身歸納出來;與逆轉6相對的是英國的法庭多了6位陪審員,而本案的陪審員中安插了幾位專家(而不都是笨蛋),這樣技術細節的精度就變得很好拿捏--這也顯出製作組的多年功力。
角色在解決萬博會殺人案的同時也開始觸及到「教授」案件,但是我腦中只剩很萌的蠟像館夫人了……另外亞雙義也果然沒死,不知道這算不算一開始就預定好的安排呢?

IV+V
炸魚薯條被強制退場,不過這在吉娜上位時幾乎已經注定了(福爾摩斯的梗)。
案件本身設計很差,疑點幾乎是隨便數都有:射擊的方向是從低到高,但是以死神的身高來說不太可能,這點前面提了一下就不了了之;不自然的死亡姿勢和即死的予盾所導出屋子不是第一犯案現場的事實應該也比腐爛的炸魚薯條和冰箱來得直接;極近距射擊造成的血痕也應該和平時不同(不只蠟燭而已);慈獄被抓上証人台沒幾下就被草草收拾掉,但如果兩次接近雪房還是個謎(別忘了只有一次20分鐘的演習)……一切都是為了推動大透情,這點令人有點失望。後期令人喜聞樂見的律師檢察官証人(?)合力坑死法官,可惜他的顏藝出現次數不足,在確定自己是一個單挑整個法庭的時候就應該多狂幾次和多拍幾次手的。暴走情景也很華麗,不過成步堂你不逃的話會被燒死的喔?

整體劇情
玩過一代的玩家幾乎所有希望看到的元素也到齊了:亞雙義和愛麗絲的身世、四個名字的電報、死神之謎、一代案件的完整詮釋:
- 引入亞雙義和愛麗絲的上一代當然有必要。一代一直強調不想讓愛麗絲知道「老爸」掛掉但是紙總會包不住火,整理她的身世十分正常;亞雙義這種有強烈使命感的人帶著「任務」去英國自然不會是見識之類的膚淺理由,唯有過去的羈絆才是唯一的解。
- 電報裡四個人「死」了兩個,不在意才怪。重點是,為何這是機密?二代的給出了不能再重的答案。
- 死神。唯有在它行動才有機會捕捉死神的影子。劇組直接讓死神退場,讓主角有機會在殘留的證據下找出死神的正體。如果有玩過逆檢1的話也能在八O鳥的退場中發現一樣的手法:要讓案件變得天衣無縫需要刑警、檢察官和律師的配合;這裡律師可以改為收買陪審團,那不聽話的班克斯要用誰來代替就很明顯了,劇組也給出了最穩妥也最喜聞樂見的答案。
- 一代完結之時就有人猜測二代可能完全顛覆一代得出的結論,有人更從一代3/5章導出所有結果均為梅根達爾操縱的結果……這個結論十分有趣(如果是真的話說不定直接變成神/糞作),可惜二代中劇組根本沒有讓與司法無關人士(証人/犯人/陪審團)操縱的機會,這樣有機會操縱的就必定是司法相關人士了--留意到這一點的人在中後期會輕鬆很多:老子就是不相信蘇格蘭場、檢察局和法官,就是覺得你們在坑我(死神好像說過一樣的說話,算是很明顯的提示了)。

可惜為了將海量的戲情塞進五章遊戲裡劇組用盡了各種方法,二三章的手法也在盡量討好傳統的逆轉粉絲群,奈何訊息量太大,被分了心的劇組在案件設計上漏洞也比過去稍多了一點。但整體來說還是保住了遊戲的完整性,尤其是與一代結合成一個完整的遊戲。

推理

難度算是中高,熟悉逆轉玩法的玩家利用消除法應該大部分時間都能比主角走快一步,剩下就是手法問題。要找出確實的證據除了要把所有可疑的物品自己調查一遍(以前都是劇情強制調查)外還要留意上文下理而不是只是紅字--紅字提示本身可能還沒有其他對話來得有用,而福爾摩斯相關的梗本身也是亦真亦假,剛好印證了他的台詞「大證探有時也會說謊的」。這樣平時當逆轉當電字小說看的玩家可能要多動一下腦子,但我覺得滿有趣的。
美中不足的是本作「逆轉」的成分真的不足。比起6代將幾個「予盾」組合逆轉得出石破天驚的結論(第三章尤其精彩),這次主角大部分時間都是順利得出結論,就算不行大家都知道福爾摩斯都有外掛讓他用,這樣大大降低了玩家找出真相的夾快感(而且犯人還不會暴走這點更是讓人不爽)。與其說是逆轉外傳不如說是像雷x逆那樣,以福爾摩斯為核心的福x逆合作遊戲。

綜觀逆轉系列

大逆轉告一段落,不知道有沒有下一集。逆檢也把大boss做了,不知道有沒有下一集。逆轉本篇王泥喜遠走他方,想寫下去只能看心音放閃。
無數證人犯人被坑,律師被坑(霧人),檢察官被坑(狩魔、夕神...),法官被坑(一下兩個),司法高層被坑(一柳、這次的Voltex)……能坑的都差不多坑過一次了。
這作算是大家想看的集大成之作,然後呢?

然後呢?

正如我提到Mario Maker時提過,如何在前作機制獲得好評時發展續作永遠是一個難題。繼續搞大新聞會不會太浮跨?加新的測謊機(外)制(掛)會不會令遊戲太複雜也喪失本格推理的特質?回歸初衷做日常案件會不會太平凡?填那個七年黑暗期的坑有可能嗎?

反正我不知道,也不需要知道。巧舟已經證明過他寫劇本的能力,之後繼續相信他就好了。

7.5/10

Wednesday 6 September 2017

Cooperative games and the FEH voting event

Consider the following situation.

Two competitors A and B, are selling homogeneous products on the market. (Well not under perfect competition anyway.) Obviously they are enemies that they don't communicate. One day, however, due to unknown reasons they decided to maximize the total profit that they create. The profit of each of the seller can be viewed as a function of the seller's price and the opponent's price, assuming that the demand curve is fixed.

The problem is that they don't communicate. They even do not bother to check the opponent's price (it's time wasting -- profit maximization is only the command from the higher order). The only information they have on their hand is their price and their corresponding profit on that day (note: it relies on the opponent's price on the day too). Is it possible to adjust the price daily in order to edge close to the goal?

Call the price at the $i^{th}$ day $P_{A,i}, P_{B,i}$ and the profit function $f(\cdot, \cdot)$ that takes seller's price and opponent's price and output the profit for the seller (assume that the market is symmetric in the sense that swapping the price results in the same total profit). Is it possible to find such iteration so that $f(P_{A,i}, P_{B,i})+f(P_{B,i}, P_{A,i}) \to L$, the theoratical maximum (such maximum exists by compactness)? Does it help if we know the smoothness of $f$?

My unproved claim: if $f$ is strictly increasing respect to both variables and is Lipschitz (why the hell do we need such assumption?) then such algorithms exists. Don't ask me for further proof...

-------------------

I do not intend to include any explicit mathematics here, but it seems like a simple example is unavoidable, so let us consider the following.

Consider two initial values $A_0, B_0$ from two players A and B. At round $n$, player $A$ will know the his own number $A_n$ as well as the absolute distance $|A_n-B_n|$, and vice versa. Without any communication we want to devise an algorithm so that $|A_n-B_n| \to 0$.

We give the algorithm as follows. For $k>1$ we iterate as follows:

$A_{n+1} = A_n + (-1)^n |A_n-B_n|/k$
$B_{n+1} = B_n + (-1)^{n+1}|A_n-B_n|/k$

then by induction the value $|A_n-B_n|$ decreases every 2 rounds, and hence it converges to zero by the monotone convergence theorem. (Well, if we take $k=2$ we get it sorted out in 2 rounds, and we get the exact solution. However convergence holds for any $k>1$, too.)

Here we used the fact that there are only 2 players, so we implemented something that has period 2. In general when there are $N$ players we want to implement algorithms that are genuinely the same on each of the players.

-------------------

The first example is definitely an interesting case to think about, but $f$ is too vague to deal with. For concrete non-trivial examples we may consider this one instead:

Consider $N$ base stations $X_1,...,X_N \in \mathbb{R}^2$, and the convex polygon created $\Omega = conv(X_1,...,X_N)$. We may assign a radius $r_i$ to each station so that the area $\Omega \cap B(X_i; r_i)$ is now covered by the station $X_i$.

Larger coverage, of course, takes more energy. The energy required to maintain radius $r_i$ is proportional to $r_i^{2+\alpha}$, where $\alpha \in [0,1]$. The goal is to find $(r_i)$ so that 100% coverage is attained in $\Omega$ and the energy consumption is minimized.

Again communication among stations is prohibited. The only information each station $X_i$ knows is the coverage rate in the area $\Omega \cap B(X_i; R_i)$, where $R_i$ is the distance between $X_i$ and the closest station. Can you devise an algorithm for stations to adjust their radius $r_i$ towards the goal? Does it become any simpler in the case $\alpha = 0$, or $\alpha = 1$?

(Note: since we have the information on coverage rates, we allow stations imperfect coverage in the mean time -- we call that beta testing -- the algorithm is fine as long as the coverage rate converges to 100% AND the energy consumption converges to the optimal number.)

-------------------

Above are games where players cooperate under limited information. These type of games are extremely useful in various fields (economics and engineering) and are also actively researched. It is important in the sense that communication is always expensive.

Given linear/convex/smooth/well-behaved functions there are systemic methods to deal with those, or at least approximation-ish result are available. But what if the 'player' is in fact a mix between a selfish self (takes adjustments to maximize sole profit) and a generous self (take adjustments to maximize total profit)? That could well happen during political elections...or game voting events. The only prediction we can make it that things are unpredictable.

That is precisely why flame arised when Camilla beat Lyn with much less 'votes' in FEH's voting event. Things are event worse in the sense that this is a zero sum game so that it is impossible to cooperate after all. People thought that they are going to tweak the result but were only dominated by the 'selfish self' - the large portion of players who simply play on their own pace and not even hardly optimizing - and when things did not go naturally in their way they went annoyed and shit everywhere...

And for me, it is no more than a sleepless currency-earning event because I did not find my favourite characters (Ursula, and perhaps Tana too) there. Hehehe.