Research on Training Events

☆☆☆ Join date : 2013-01-06 Posts : 330

New training event tomorrow! (Mina kills the gardeners)

Keep the data flowing. Need those 6* data points.

Please read the first post for instructions on how to post data. It will make it much easier for me, so please and thank you!

☆☆☆ Join date : 2013-01-06 Posts : 330

DATA

Reserved.

☆☆☆☆☆ Join date : 2012-09-24 Posts : 954

I wonder if I should post the answers to this...hmm

☆☆☆ Join date : 2013-01-06 Posts : 330

How much data do you have? Last I saw before I started this, it was about as much as I have now, which clearly isn't enough.

I'd love to see what you have Wink

.

☆ Join date : 2013-02-20 Posts : 70

DATA

As I play more, I'll try to keep recording my information here and update the numbers below. Let me know if you want any other kind of information in these posts.

**FINAL RESULTS BEFORE MAY 1ST UPDATE

5* Siever Lead (non-event)

Training 26-5

Spoiler:

Training 1-5

Spoiler:

Training 11-5

Spoiler:

Random Training While Trekking to Higher Levels

Spoiler:

TOTAL
29770 TP
4* Encounters: 278
5* Encounters: 14
6* Encounters: 5
King Mandragora Encounters: 2

☆☆☆ Join date : 2013-01-06 Posts : 330

Looks good! Thanks for helping.

☆☆☆ Join date : 2012-09-29 Posts : 456

Make sure to be careful about fever time skewing results.
As well as King Mandrake/Mangragora/whatever it's called

☆☆☆ Join date : 2013-01-06 Posts : 330

Oh yeah, I noticed the banner said that it also increased chances of rare bosses. This will complicate things...

Mandrake is easy, I'll just stick him in his own category as I DON'T think he has the same encounter rate as other 4*s.

☆☆☆☆☆ Join date : 2012-09-24 Posts : 954

Mmmm, to give the data or not...
There are quite a few people in this game I would rather not give the info to...
And posting it on the forums 'is' akin to giving it to them...

☆☆☆☆ Join date : 2013-02-24 Posts : 748

Well, my playing style is quite hard to record the data, so I cannot give actual data for this.
But thanks for all the hard work.

☆☆☆ Join date : 2013-01-06 Posts : 330

For those looking at the Spreadsheet... The current event is in the second tab. I needed to segregate it to account for fever and the king.

Anecdotal (and very preliminary) evidence suggests that the current event has tweaked quite a few things, from encounter rate to the percentage of those being rare encounters, so I'm not sure how much "help" BK's data would give for the current event or future events for that matter. Either way, I'll keep this up and running. I won't do sweeps as often as I did before though. Maybe once every couple of days.

☆ Join date : 2013-02-20 Posts : 70

Wait, where's the second tab?

☆☆☆ Join date : 2013-01-06 Posts : 330

At the bottom Wink

.

☆ Join date : 2013-02-20 Posts : 70

Oh, there it is! Thanks I love you

☆☆☆ Join date : 2012-10-23 Posts : 481

errrrm, what makes me wonder is that with the current data in spreadsheet it is actually shown that 1-5 is a better stage to train than 26-5...

☆☆☆ Join date : 2013-01-06 Posts : 330

BadEndNight wrote:: errrrm, what makes me wonder is that with the current data in spreadsheet it is actually shown that 1-5 is a better stage to train than 26-5...

Are you looking at the right tab? For the old training events that was true, but for this one, the data shows the same encounter rate per TP used for both. My data is a little lower than Twilight's, but I have a brand new set to throw in tonight that will definitely increase it.

Also, I have been working under the assumption that Fever Time does NOT increase actual encounter rate, just rare encounter rate... but I'm starting to think that might not be the case... wonder if I should start to separate those two set of data.

☆☆☆ Join date : 2012-10-23 Posts : 481

Sabin76 wrote:

BadEndNight wrote:: errrrm, what makes me wonder is that with the current data in spreadsheet it is actually shown that 1-5 is a better stage to train than 26-5...

Are you looking at the right tab? For the old training events that was true, but for this one, the data shows the same encounter rate per TP used for both. My data is a little lower than Twilight's, but I have a brand new set to throw in tonight that will definitely increase it.

Also, I have been working under the assumption that Fever Time does NOT increase actual encounter rate, just rare encounter rate... but I'm starting to think that might not be the case... wonder if I should start to separate those two set of data.

I think I'm looking at the right tab with the same encounter rate per TP as you mentioned. However, you can't simply rely on encounter rate per TP in mathematics strictly speaking.

Here's an simulation in computer, showing (encounter in 28 1-1 run - encounter in 10 26-5 run) which is simulated a million times which should be extremely close to where the truth should be.

Code in R, if you want:

Spoiler:

Research on Training Events - Page 3 2InJJ

That says, in 47% of the time 1-1 performs better (with 18% for a draw). In average, you expect 0.4 encounters more in 1-1 per 280TP spent.

☆☆☆ Join date : 2013-01-06 Posts : 330

BadEndNight wrote:

Sabin76 wrote:

BadEndNight wrote:: errrrm, what makes me wonder is that with the current data in spreadsheet it is actually shown that 1-5 is a better stage to train than 26-5...

Are you looking at the right tab? For the old training events that was true, but for this one, the data shows the same encounter rate per TP used for both. My data is a little lower than Twilight's, but I have a brand new set to throw in tonight that will definitely increase it.

Also, I have been working under the assumption that Fever Time does NOT increase actual encounter rate, just rare encounter rate... but I'm starting to think that might not be the case... wonder if I should start to separate those two set of data.

I think I'm looking at the right tab with the same encounter rate per TP as you mentioned. However, you can't simply rely on encounter rate per TP in mathematics strictly speaking.

Here's an simulation in computer, showing (encounter in 28 1-1 run - encounter in 10 26-5 run) which is simulated a million times which should be extremely close to where the truth should be.

Code in R, if you want:

Spoiler:

That says, in 47% of the time 1-1 performs better (with 18% for a draw). In average, you expect 0.4 encounters more in 1-1 per 280TP spent.

Well, you definitely didn't use the wrong data. A couple things: The fact that chapter 1 already showed a slight statistical advantage in encounters/TP was not lost on me (.0112 to .0110). I simply saw a vast deviation from the previous formula and am assuming that they will trend to the same value over time. I would ask that you run the simulation again using percentages that would have identical encounter/TP ratios. If it still shows an advantage for chapter 1 (which it shouldn't), then I would say your simulation has merit. On the other hand by your simulation 35% of the time 26-5 is better, and with the addition of being able to get 2* and 3* cards to make TE off of, I'd still say 26-5 has the advantage Wink

.

Second, the data is still rather preliminary. As I mentioned, I based my conclusion off of a trend I saw in the data, not the (relatively few) numbers I actually had. We will all have to wait and see what the data says by the end of the event. As I get more people reporting other chapters, we may yet find something different.

☆☆☆ Join date : 2012-10-23 Posts : 481

Changing the parameter from 0.1123 to 0.11 and let's skip to the conclusion:

> mean(diff) #0.395096 (<---- original value)
[1] 0.329557
> var(diff) #4.771424
[1] 4.729194
> sum(diff>0)/n #0.472596
[1] 0.460088
> sum(diff==0)/n #0.181848
[1] 0.183946

That says, 1-1 still has 0.33 encounter more on average, and is not worse than 26-5 in 64% of the time. The curve plotted is basiclly the same with the previous one, shifting to the left with a magnitude that no likely to be noticed.

I am also interested to do a regression on mandora VS level of boss fixing rarity but the variation is strange and the regression is barely conclusive.

☆☆☆ Join date : 2013-01-06 Posts : 330

BadEndNight wrote:: Changing the parameter from 0.1123 to 0.11 and let's skip to the conclusion:

> mean(diff) #0.395096 (<---- original value)
[1] 0.329557
> var(diff) #4.771424
[1] 4.729194
> sum(diff>0)/n #0.472596
[1] 0.460088
> sum(diff==0)/n #0.181848
[1] 0.183946

That says, 1-1 still has 0.33 encounter more on average, and is not worse than 26-5 in 64% of the time. The curve plotted is basiclly the same with the previous one, shifting to the left with a magnitude that no likely to be noticed.

I am also interested to do a regression on mandora VS level of boss fixing rarity but the variation is strange and the regression is barely conclusive.

Well, there it is. We'll have to see what the numbers look like at the end, but so far, you're right. Now this has me thinking... Somewhere in the back of my mind this actually makes sense, but I'm at a loss to explain why.

Can you PM me? No reason to bore the general population with this Wink

.

P.S. My inbox hates you.

☆☆☆ Join date : 2012-10-23 Posts : 481

I think another important point is to compare mandaro obtained instead of encountering rate.

As hypothesized rarer boss appears more in later stage (esp. the king), this plays an important role in optimization.

☆☆☆ Join date : 2013-01-06 Posts : 330

True, though I hadn't been tracking that. I would advise against going with the anecdotal evidence in the event thread, however. Observer bias runs rampant.

Also, I'm noticing that my personal numbers are lower than those reported by other players in both events. I'm wondering if there's a discrepancy in how people are recording their data.

I do this:
I count clicks that take TP away (ignoring player meetings) until I either run out of TP or encounter a boss. If I run out of TP, I record the number of clicks and put an "X" next to it. If I encounter a boss, I put the number of clicks and then the rarity of the boss down. When I put the data in my spreadsheet, I just multiply the number of clicks by the TP/click for the training I was in. This way I'm recording actual TP spent, instead of full bars used, etc. TP doesn't regenerate when you are actively training, but it DOES regenerate when you are in a battle, so doing it the latter way will show less TP spent than what the actual value is.

Subject: Re: Research on Training Events Sun Apr 28, 2013 4:54 pm

DATA

Leader: 6* Aksana
Training Episode : 26-5

Total TP used: 2800 (100 advances)

Fever Times:

1: After 4th advance
2: After 1st advance
3: After 1st advance
4: None
5: After 1st advance
6: None
7: After 1st advance
8: None
9: After 6th advance
10: After 2nd advance

4* : 30
5*: 1
6*: 0
Mandragora King : 1 ( Was under Fever Time )
Mandragora Boxes (wasn't sure to include this?) : 5

It's kinda weird imo, before writing down all this I was using Adele as my leader and ran into 5 Agastaches but now I barely ran into 1 :/. Will update this soon with Other episodes using 100 advances.

Side note : Next time I'll include the monsters I run into during fever and not during fever. Totally forgot about that, my bad u.u

☆☆☆ Join date : 2013-01-06 Posts : 330

Thanks Tash. Be sure you aren't counting when you meet another player in training as that doesn't use up your TP.

☆ Join date : 2013-02-20 Posts : 70

DATA

**Updated

Training 1-5

Spoiler:

TOTAL
14750TP
4* Encounters: 140
5* Encounters: 6
6* Encounters: 1
King Mandragora Encounters: 0
Mandrake Chests: 28

» How to grind for the 7★ shard/pin/ect. in training events?
» Sea Paradise Training: What is best training level to grind?
» Got questions? Ask here!
» Clash research - values on reaching the completion 8*
» The Research Lab [Working List Posted, PLEASE READ!]