[POE] Building Better Builds: Forum Posts, Part i

I've become interested in making my Path of Exile builds better.

One way to do this is to see how other players are creating their builds. Since this is a pretty big topic, I'll first focus on passive skill tree choices by players.  I'm particularly interested in these because many players post guides to their builds - including passive skill trees - on the forums. (Data time!)

I grabbed the 600 most recently updated threads in each class forum and extracted build links on the first page of each of these threads. Because PoE's passive skill tree is constantly changing, I removed builds that are no longer valid with the current skilltree or were less than 80 points. Because sometimes authors post multiple versions of a build, I selected only one build from each thread (the build closest to 100 points). This left me with 914 valid builds.

	Duelist	Marauder	Ranger	Scion	Shadow	Templar	Witch
valid builds	101.0	106.0	144.0	123.0	166.0	112.0	162.0
all builds	707.0	540.0	657.0	732.0	633.0	591.0	615.0
average posts	68.8	79.3	47.7	68.2	49.1	58.8	80.6
average views	25071.7	33006.9	17866.2	18892.1	14617.9	15087.8	32962.1

Curiously, Templar has the least build options to chose from, combined with relatively few posts / views for each thread. I'm curious whether this will lead the developers to consider rebuilding the templar starting area.

Defensive specialization: 
Whether in a hardcore or softcore league, one of the biggest challenges in Path of Exile is simply staying alive. For this section, I'll focus on player choices to get life, energy shield, armor, and evasion. When evaluating life- or energy- based builds, I omitted any build with less than 50 of that stat (because, well, death).

A few observations:

• Life-based builds tend to get around 200 life, with marauder builds getting slightly more life than the other classes. I'm actually surprised at how many builds get less than 200 life, which I considered an absolute necessity. 
• Rangers and shadows tend to focus on evasion, while duelists and marauders focus on armor. 
• Of all the classes, only witch has multiple builds getting a 'reasonable' amount of energy shield for CI / low-life. 

There is, of course, a potential problem with this data. The wide range of life / energy shield values suggests some of these builds simply aren't viable. For example, a large number of builds from new or inexperienced players might tend to undervalue defensive stats. To correct for this, I weighted the data by the number of posts in the thread (on the assumption that high-quality builds are more likely to get attention).

While weighting the data did remove a large number of the low-defense builds, the basic story is unchanged: players tend to get at least 200 life, energy-shield based builds are uncommon, and strength-focused classes get armor while dex-focused classes get evasion. It is notable that a few high-attention builds (like cyclone marauder) tend to dominate the data.

Weapon specialization:
Turning to offense, here I looked at whether a build grabbed a damage specialization node (+damage with a weapon type / spell).

A few features that stand out (to me):

• Wow, caster options are popular (even omitting elemental damage builds).
• Ranger is overwhelmingly the most popular choice for bow builds. The lack of a bow-templar makes me consider running one.
• One-handed builds are more popular than two-handed builds for all classes.
• Swords are slightly more popular than axes, despite the focus on soultaker in the current meta, though this is potentially due to buzzsaw builds. 
• Specializing into wands / staves are remarkably uncommon options (2.6% and 1.4% of builds, respectively). Staves make sense - other than Pillar builds, not many characters go for a staff (they're also devilishly hard to roll). But wands? I seem to remember everyone trying wander early in the league.

Thoughts:
I'm frankly a bit surprised at how few defenses many builds grab, particularly given the common refrain that 200% life is a bare minimum for PoE. I'm equally surprised at how many builds choose to specialize in a weapon, given how that limits purchasing options at endgame.

There are plenty of places I'm considering taking the analysis. I'm definitely going to look at keystones. I'm also interested in some general descriptive-stats for the passive tree (shortest paths, nearest clusters) and how this influences the build meta.

So many fun data adventures to be had.

[GW2] Server Transfers

I've been tracking players on the GW2 achievement leaderboards.

The boards enable me to track when a player transfers off of a server: players that disappear from the boards - but have enough points to appear on them - have likely jumped ship.  (Or, alternatively, been banned: I'm not sure what happens with the boards then.)

I was able to track players on 23 EU servers for one month [February 20th to March 23rd]. In that month, I observed 1456 players that likely transferred servers. That makes a 0.2% chance that a top 1000 player server transfers on any given day - or - a 6.6% chance that a top 1000 player changes servers over the course of a month.

Caveat: some of my data collection was corrupted - from both my ISP (TimeWarner <glare>) and the ArenaNet website (the website got angst-y partway through my data collection). So I'm missing some servers. Take the resulting numbers with salt.

Transfers are highly clustered: on most days on most servers, at most one player out of the top 1000 transfers (65% of all server-days look like this). However, on some servers, on some days, large clusters of players leave the server (record observed: 57 of the top 1000 players left Underworld on March 20th).

Transfers also differ by day: on March 7th, ArenaNet announced a reduction in the cost of server transfers starting on March 18th.  This resulted in a considerable rise in the number of transfers. Before March 18th, an average of 1.9 top 1000 players transferred per server each day. After March 19, that number jumped to 4.7.

	Pre March 18	Post March 18
Gold	1.1	1.4
Silver	2.0	9.2
Bronze	2.4	4.8

But while these transfers were designed to encourage players to move to lower-tiered servers, they had something of a bandwagon effect.  After the update, very few top 1000 players left gold-tier servers. Most of the movement came from the Silver / Bronze tiers.

Thoughts
I'm actually surprised at just how common server transfers are. Post-March 18, nearly 10 of the top 1000 players in Silver tier on every server changed servers each day. That is a remarkable change in the population of each server.

I'm also not surprised that most transfers originate on Silver / Bronze servers.  WvW match-ups on these servers can be frustrating, and it makes sense that players would seek a stronger world environment.

[GW2] Server Populations and Achievement Scores

After observing large differences in Marionette success rates across servers, I've become interested in understanding a bit more about server populations.

So the achievement point leaderboards were of particular interest to me.

Players earn achievements by completing milestones, participating in living story events, and completing dailies. They're a great measure of player activity, since you get achievement points for just about everything. Even better, ArenaNet publicly tracks achievement points for the top 1000 players on each server.

Here, I've plotted the achievement points of the top 1000 players on each server.  The servers are organized by the average achievement point score of these players (highest top left to lowest bottom right), and colored by their WvW season 1 tier (gold / silver / bronze).

There are vast differences between servers.

As a snapshot, the leaderboards are interesting.  But tracked over time, they contain much more information. For this, I collected data from the leaderboards during the Edge of the Mists update. This gives three measures of server activity:  (1) transfers, (2) inactive players, and (3) points gained among the active players.

Server Transfers: Jumping Ship
Server transfers show up on the leaderboards in two ways. A player could appear on one server, and later on another. So I can track accounts that move between servers.

Other players disappeared from the leaderboards. So, for example, if a player on Eredon Terrace had over 4217 points at the start of the Edge of the Mists update, they should appear on that server's top 1000 list unless they transferred off the server.

Overall, 1.6% of players listed on the leaderboards transferred servers. Given the small number of transfers from each server (and the reasonable expectation that some of the 'large' moves are due to guilds) it is hard to make a general pattern from the server transfer data, but I'll look at it again in future updates.

However, of the transfers I observed, 84% are to servers with higher average achievement point scores than the starting server.

Inactive Players
Overall, 17% of players on the server top 1000 didn't earn any achievement points during the Edge of the Mists update, and therefore likely didn't play GW2 during the update. Here are the percent of inactive players by server:

Inactive players are more common in lower-tiered servers, with the Bronze tier averaging over 23% of players inactive, compared to an average 12% for the gold servers.

Points Gained by Server
During the two-week update, active players on the leaderboards gained an average of 170 points, around 12 points per day. Pretty impressive, considering that the daily requires only 5 points. (The average top 1000 player earned about one-third of the total possible points to be earned from living story / daily content.)

Across all servers, players with more achievement points earned more points than those with fewer points (rich get richer faster).

Several Thoughts
By now, it is clear that there are large differences in server populations and activity. It is simply a myth that there are 'pve servers' that just happen to do poorly in WvW. Players tend to transfer to higher-tiered servers. Those servers have fewer inactive players. And the players on higher tiered servers tend to be more active.

Time to think about a server transfer myself...

Sidenote: the results presented here don't change under more complicated statistical models (which I ran, but excluded from the presentation because this is a long post already). Though, I did have to run a nonparametric model due to the interesting structure underlying the data.

[GW2] Professional Voting

Today, Guild Wars 2 extended their Collaborative Development Initiative (CDI) to ask players to post their feedback on the ranger profession.  (Note: 'professions' are what most MMOs would classes).

Ranger was chosen in this thread, where the developers asked players to list 3 professions (in order of priority) they would like to see discussed first.  But isn't choosing a profession based on posting in a thread open for 24 hours on the forum problematic? Particularly active forum users would be likely to overwhelm the thread. And its reasonable to expect that posters would be biased toward their own class, instead of looking out for game balance as a whole.

So how reasonable was the thread as a method of getting feedback?

The results:

Ranger was the most mentioned profession, followed by elementalist and then necromancer to round out the top three. Rangers are particularly weak in GW2's pve, and one playstyle (bearbow) is consistently viewed with scorn.

Breaking down the votes by order of priority indicates that not only was ranger the overall top choice, it was most frequently listed as priority #1 by voters. Most of the ordering between professions remains the same.

But are these votes contaminated by bias?

Prior Posting Activity

One objection to this poll might be that not only are forums generally unrepresentative of player opinions, but frequent forum posters might be likely to overwhelm any thread. Hardcore and casual players might be expected to disagree on which professions need the most attention.

While generally valid objections, they don't apply particularly well to this poll.

For nearly 20% of voters, voting in this thread was their first post on the forums. For others, posting is a pretty frequent activity. Voters had an average 205 posts before the thread (with a maximum of over 4600 posts!). 

Curiously, however, prior posting activity doesn't change the results. Both first-time and prior posters overwhelmingly agree on their preference order, favoring rangers, elementalists and necromancers. (Though mesmers do a little less well among first-time posters).

Profession(al) Bias

Another way to look at the votes is to examine what other forums voters are active on. Players might be expected to be biased toward voting for their own profession, which could be reflected by their posting in profession-specific subforums.

Here, I took a look at the past 50 forum posts of each voter (excluding first-time posters), and recorded whether they'd posted in one of the profession-specific subforums.  I compared this against their listed vote preferences.

In general, the votes are only slightly biased toward forums voters are active in.  For example, 30% of those who ranked elementalist as priority #2 had posted in the elementalist subforums, compared to only 15% of all voters. But this pattern doesn't hold generally. Those who picked engineer, mesmer, thief or warrior were no more likely to be active in those professions' subforums. So, in general, forum posting activity is at best weakly related to vote choice.

The exception - as ever - is rangers.  While only 5% of voters had posted in the ranger forums previously, 45% of those who listed ranger as their top pick had posted in those forums. Among top choices, ranger is the only profession that shows such a strong pattern. However, even excluding these voters, ranger would still be the top pick.

Thoughts

Clearly, a 24-hour poll isn't the best way to generate clear feedback on which profession requires the most attention.  In this case, however, the ranger profession is so overwhelmingly in need of attention that the effect of these (otherwise valid) concerns is overwhelmed.

Fingers crossed for some pretty excellent fixes.

Disclaimer: I have a ranger at 80, but I only use it to loot jumping puzzle chests since returning to GW2.

[GW2] Data Mining and Magic Find

I've been busy data mining in GW2.

This time, I was curious whether magic find affected the chance of getting gems from mining nodes.*

I've seen plenty of questions about magic find and gathering. I've seen speculation on the topic. I've seen assertions given (many assertions). But the wiki isn't much help. And I haven't seen evidence.

So I set about mining.

Every day for the past two weeks, I took each* of my level 80 characters around Tyria gathering orichalcum and any mithril along the way. All characters used orichalcum mining picks and had the guild gathering bonus. One character had 101% MF, the others only had the guild magic find bonus (10%). By the end, I gathered 1027 mithril and 1483 orichalcum.**

Data can be found here.

All told, I didn't see a difference in gems between different levels of magic find.

Node        MF     N    Gem Chance
Mithril     10%:  696   9.5% (1.1)
Mithril    101%:  331   9.4% (1.6)
Orichalcum  10%: 1065  12.0% (1.0)
Orichalcum 101%:  418  12.7% (1.6)

While orichalcum nodes have a higher chance of yielding gems than mithril nodes, I see no difference in the chance of getting gems between 10% MF and 101% MF. (For the stats-nerds, checked with a probit.)

Of course... this is a relatively small sample. So how likely is it that magic find affects the chance of getting gems from mining, but I just didn't notice it? (In stats-language, what is the power of the test, given expectations about the size of the effect being studied.)

To check this, I simulated the results of gathering between 70 to 4200 mining nodes (increments of 70), assuming that magic find works multiplicatively. For each simulation, I checked whether I found that magic find had an effect. (I also included differences between node types, and recorded those.)

From the plot, its clear that after around 1500 mining attempts I'm nearly certain to find an effect from magic find, if one exists. At about 2500 mining attempts (the data I've collected), assuming magic find works, I'd have less than a 1% chance of observing no effect.

TLDR: magic find doesn't affect the chance of getting gems while mining.

But, of course, I'll keep gathering data, and check back in if any results listed here change. Because science is the joy of revision.

Related links:
On this blog: MF doesn't affect bags.
From reddit: MF doesn't affect salvaging.
Dev: MF only works on kills.

* Horray for leveling via crafting. I love going from 10 to 70 in like 2 hours.
** Just to be safe, this data excludes any ruined ore / gems from ruined ore. Which is why neither total is divisible by 3.