josh_nc
Member
 Offline
Activity: 95
Merit: 10
|
 |
May 20, 2013, 04:11:16 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).  |
|
|
|
|
ProfMac
Legendary
Offline
Activity: 1246
Merit: 1002
|
|
May 20, 2013, 04:19:53 AM |
|
Again, no. Sorry! This is all very tricky and I suggest you read about non-homogenous Poisson processes and the geometric distribution. Variance will increase as the hashrate increases, but it will remain a nearly constant proportion of the hashrate.
Now you're just throwing crazy words out there left and right! Do you realize how much time it would take someone to google all these words? - non-homogenous
- Poisson processes
- geometric distribution
- Variance
 Is there any reason you chose the most extreme values in the red graph, when you also had the choice of some more moderate values in the green graph, when you were trying to understand what was happening? Neither of them should be used. Did you quote the wrong person? Yeah. I meant josh_nc. I'm more interested in why he made his choices, vis-a-vie interpreting such random things as "the hockey stick" than I am a high finess resolution of which model is best. |
I try to be respectful and informed.
|
|
|
|
|
organofcorti
Donator
Legendary
Offline
Activity: 2058
Merit: 1007
Poor impulse control.
|
|
May 20, 2013, 04:30:21 AM |
|
So, would you say the approach below would be a better way to determine the hash rate? This of course takes into account the maximum likelihood. L() is in fact the likelihood function. I've always been terrible with statistical models but this one may give you a more coherent distribution than Poisson's which smells really fishy wink wink   Can you explain how the above likelihood function would help? I'm not sure I follow. Here's how I understand estimates of the network hashrate: Firstly, if the network hashrate is unchanging then the number of blocks per unit time is Poisson distributed. Any hash has the same chance of being sufficiently under target to solve a block, the number of hashes per block is geometrically distributed, with p = 1/target. Since p is extremely small we can approximate this quite closely using the exponential distribution. Assuming the hashrate doesn't change, then the amount of time between block solves is exponentially distributed, and thus the block solving process is a Poisson process. However the block solved rate changes, sometimes significantly, between any two retargets. This makes the process a non-homogenous Poisson process. So you can calculate the hashrate from the number of blocks per time, or the inverse of the amount time per block/s, and use confidence intervals based on whichever method you used (Poisson distribution for blocks per time, Erlang distribution for amount time per block/s). My own method is to use a non-parametric smoothing spline. I then time stretch the data so that the smoothing spline is 12 blocks every two hours and optimize such that all of the actual two hourly hashrate are Poisson distributed random variables with a mean rate of 12. |
|
|
|
josh_nc
Member
Offline
Activity: 95
Merit: 10
|
|
May 20, 2013, 04:32:47 AM |
|
Geek humor. Never mind any of that. Are you posting your graphs somewhere? |
|
|
|
|
organofcorti
Donator
Legendary
Offline
Activity: 2058
Merit: 1007
Poor impulse control.
|
|
May 20, 2013, 04:39:02 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful. Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off. What you say might be possible, but it is by no means even close to being provable. |
|
|
|
organofcorti
Donator
Legendary
Offline
Activity: 2058
Merit: 1007
Poor impulse control.
|
|
May 20, 2013, 04:41:21 AM |
|
Geek humor. Never mind any of that. Are you posting your graphs somewhere? Are you saying I'm not geek enough? I want satisfaction, sir! organofcorti.blogspot.com is where I post most of my odds and ends. Unfortunately I don't have sufficient web-fu to make a website that would serve a real time network hashrate chart though. |
|
|
|
josh_nc
Member
Offline
Activity: 95
Merit: 10
|
|
May 20, 2013, 04:43:51 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful.
Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off.
What you say might be possible, but it is by no means even close to being provable.
It could be a combination of ASICs coming offline (Avalon mining with batch#2 before shipping them on the 13th), then online (people receiving them and turning them on) and all at the same time FPGAs leaving the pools. Long shot for an explanation but something seems the have been going on in that period. |
|
|
|
|
ProfMac
Legendary
Offline
Activity: 1246
Merit: 1002
|
|
May 20, 2013, 04:45:41 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful. Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off. What you say might be possible, but it is by no means even close to being provable. It could be a combination of ASICs coming offline (Avalon mining with batch#2 before shipping them on the 13th), then online (people receiving them and turning them on) and all at the same time FPGAs leaving the pools. Long shot for an explanation but something seems the have been going on in that period. People mining alt coins? |
I try to be respectful and informed.
|
|
|
Minor Miner
Legendary
Offline
Activity: 2464
Merit: 1020
Be A Digital Miner
|
|
May 20, 2013, 04:48:05 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful. Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off. What you say might be possible, but it is by no means even close to being provable. I do not understand why three days is not enough to account for the variance of block solving times. It is 336 events no? Some will be tight together pushing the implied HR higher but some will be further apart? I could see it being skewed by 30 or 40 events but I am shocked that over 300 is not enough to predict. So, the chart on bitcoincharts is useless? |
|
|
|
josh_nc
Member
Offline
Activity: 95
Merit: 10
|
|
May 20, 2013, 04:52:19 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful. Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off. What you say might be possible, but it is by no means even close to being provable. It could be a combination of ASICs coming offline (Avalon mining with batch#2 before shipping them on the 13th), then online (people receiving them and turning them on) and all at the same time FPGAs leaving the pools. Long shot for an explanation but something seems the have been going on in that period. People mining alt coins? Alt coins don't pay and FPGAs are no longer worth it since ASICS. Check this list of all available crypto currencies and the calculator below. Plug in some typical values for alt coins (scrypt) and you will see what I mean. Keep in mind that for scrypt the top miners are in the 1-4 MHash/sec: http://dustcoin.com/mining |
|
|
|
|
|
BeetcoinScummer
|
|
May 20, 2013, 06:13:01 AM |
|
Alt coins don't pay and FPGAs are no longer worth it since ASICS.
It is no longer cost-effective to BUY FPGAs. Running them is still OK. Even the most power inefficient FPGA (the BFL single) is still worth it to run. For me it still makes around 10x the electricity it uses and my electric rates are pretty high. |
|
|
|
|
organofcorti
Donator
Legendary
Offline
Activity: 2058
Merit: 1007
Poor impulse control.
|
|
May 20, 2013, 06:28:56 AM |
|
I believe this would be a better graph to look at. The purple line is the 3-day estimate. For the period I brought up (May 13-16) it is a lot more coherent. The slow decline bet. May 14 and May 19 could be the FPGAs being turned off. Pools like BTC Guild are kicking them all out at this point because they cannot use the stratum servers and they want to switch off their getwork servers (old stuff).
[ ... ]
Even a three day estimate is too short. The 14 day window estimate is more useful. Thing is, that to create such massive changes in the network you need a large number of people in different timezones to act at exactly the same time. That drop looks to be about 10 Thps - it's more likely that this drop is due at least in part to variance, rather than 12500 800Mhps FPGAs being switched off. What you say might be possible, but it is by no means even close to being provable. I do not understand why three days is not enough to account for the variance of block solving times. It is 336 events no? Some will be tight together pushing the implied HR higher but some will be further apart? I could see it being skewed by 30 or 40 events but I am shocked that over 300 is not enough to predict. It's not that it won't be accurate, just that there's no confidence interval for it's use. It might be correct, or it might not. The 14 day estimate is closer to my own estimate, and my own estimate is provably a good fit for a time streched Poisson process (essentially a non-homogenous Poisson process transformed using a variable timescale so it becomes a Poisson process). It also allows confidence intervals. The 3 day estimate is affected far more by variance than my own, and to me the estimate doesn't seem like it would be a good estimate for the mean. If I could access the data set I could find out for sure. So, the chart on bitcoincharts is useless?
Depends on what you're using it for. If you want to know the hashrate, then yes, useless. If you want to know how lucky the network has been recently, then it's probably a reasonable indicator. |
|
|
|
josh_nc
Member
Offline
Activity: 95
Merit: 10
|
|
May 20, 2013, 08:48:49 AM |
|
Alt coins don't pay and FPGAs are no longer worth it since ASICS.
It is no longer cost-effective to BUY FPGAs. Running them is still OK. Even the most power inefficient FPGA (the BFL single) is still worth it to run. For me it still makes around 10x the electricity it uses and my electric rates are pretty high. You may want to check into mining Terracoins. It seems to be going into a boom right now and you should get 2.4 times more than with btc. Apparently (just found all this out), fpga miners are all flocking to trc. |
|
|
|
|
|
|
|
Vicus
|
|
May 20, 2013, 09:25:19 AM |
|
Anyone have or can make shorthand report of this video's? Too much talker echo, difficult to understand for foreign listener. |
|
|
|
|
mtbitcoin
Legendary
Offline
Activity: 876
Merit: 1000
Etherscan.io
|
|
May 20, 2013, 10:06:34 AM |
|
That was a good video, thanks for posting the link (or whoever the original author) was. Cleared up a lot of stuff (about the reason for them not answering tickets and all) and Yifu does seem like a real decent guy. Hopefully this means we should have our units within the next few weeks  |
|
|
|
BitAddict
Legendary
Offline
Activity: 1190
Merit: 1001
|
|
May 20, 2013, 10:11:21 AM |
|
Anyone have or can make shorthand report of this video's? Too much talker echo, difficult to understand for foreign listener. +1. report please  |
|
|
|
|
riqpit
Newbie
Offline
Activity: 30
Merit: 0
|
|
May 20, 2013, 11:02:21 AM |
|
thanks for the link / video. any ideas about the rest of Q&A - maybe parts 2 & 3? |
|
|
|
|
SolarSilver
Legendary
Offline
Activity: 1112
Merit: 1000
|
|
May 20, 2013, 11:59:00 AM |
|
any ideas about the rest of Q&A - maybe parts 2 & 3?
I did not make the video or upload it, I just searched for it. If you look at the digitalmagus7 history, he posted a whole bunch of Bitcoin 2013 video, perhaps more is to follow. https://www.youtube.com/user/digitalmagus7/videosAs for the info in the video, I have a hard time transcribing it but 4:12 "We are just going to sell chips" (then at least they don't have to deal with customers, shipping, customs) 4:40 "A slight sabotage issue" (about the units that arrived broken, more at 9:08) 5:10 "And after we double check our new SMT line production bla bla in a few weeks from now, ship them out in pockets" 6:20 "Q: Are you mining with batch #2 or not mining? We are not mining" So no explanation why they missed batch #2 shipping date, no explanation why some units arrived 'used'. No fixed shipping date for batch #2, except for "in a few weeks" So... June? |
|
|
|
|
|
