The KDB issue (i.e. rising whilst -ve) was due to wild cards in late October. Similar issues have occurred before at this stage of the season but not with such a dramatic outcome. I can see what needs to be done to mitigate future events. However it is not a quick fix.
Looks like we are in the midst of a wild card onslaught, not yet a tsunami but a still a sizeable flood.
All test cases passed last night - less programming adjustments - Yippee!!!
A quick review of Son's history indicates that the most likely cause for last nights prediction error of 44% are i) the excessive volume of wild cards played during game week 4; and ii) the low transfer rates during the current international break.
Richarlison is a 'test case' for tonight. If he drops (and isn't predicted to drop of course) then I will know to change the rule in the morning. Just letting everyone know in case transferring him out is important to your budget.
Wow!!!
Maddison first big error of the season. (Prediction 57.4%). Can see the reasons. He should have been set at 86.8% Adding in the missing % increment was delayed due to his injury (Oops!!!!!) The rest of the error is due to flow reversals as he was previously close to -100%
Could the person on IP Address 3.250.158.99 please stop hacking the website. They are running a python tool that is set to repeat every 10 seconds. If this does not cease within the 30 minutes I will submit a a report to the relevant authorities.
The automated calibration for the rises went well last night. For the drops less so, roughly a 9% error.
Issues are: i) the wild cards are picking up pace (this is the norm at this stage of the season) and ii) one of back-end systems is misbehaving (will address later tonight).
The top 50 managers have played 87 chips (BB 16, FH 12, TC 36 and WC 23). 16 chips were played during the last game week. Only 1 of the top 50 teams have not played a chip. Clearly at this time the vast majority of the top 50 are not serious long term title contenders.
A wild card onslaught is gaining momentum tonight. The machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s. Therefore expect some overly enthusiastic predictions and a few prediction inaccuracies.
A wild card onslaught is currently gaining momentum, but has not yet reached last weeks 'mega' proportions. Hopefully it's just a small wave. However, do expect overly enthusiastic predictions and the occasional sizeable error whilst the wild card volumes are high.
Bamford was the first sizable error of the season. An error of 13.1%. The reason (a buglet in some coding) is obvious (they always are after the event). He should have been set at <-100%, i.e. predicted to drop. I will now correct the coding to avoid a repeat embarrassment.
The offending two tweets have now been deleted. They were regarding Saka (which was 100% accurate) and Firmino (5.2% error). I suggested he needed glasses and anger management classes.
1,117,718 wild cards were played during the international break. I'm somewhat surprised that the site didn't get completely inaccurate during this period. But it does explain why a number of players were refusing to rise / drop!!!
There appear to be an increasing number of posts regarding conspiracy theories concerning the price rises and drops. I thought I should make it clear that I have found no evidence to support these claims. Any prediction errors are typically due to a lack of data or programming.
Still no evidence to support any manipulation or conspiracy theories regarding price changes. E.g. Vardy was borderline last night and only passed 100% late on.
My only concern is with Wanyama's price drop. Although close at -96.8%, it violated a statistical trends on the site
A wild card onslaught is gaining momentum. Expect overly enthusiastic predictions whilst the volumes are high. Whenever the total volume of WC transfers for a player exceeds the effective transfers for that player, then predictions for that player become extremely difficult
A tsunami of wild card activations is underway. This is the norm as the int break gets underway. Expect some overly enthusiastic predictions. Machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s. Hopefully calmer later.
Yippee - Found a bug in the old code when testing a new programme component installed yesterday. This will improve the drops. Only problem is that it needs a re-calibration first. So no benefit tonight.
ALERT: Probably just tweeting the obvious. Over this Game Week we are experiencing an onslaught of 'Wild Cards'. Whenever the total volume of WC transfers for a player exceeds the effective transfers for that player, then predictions for that player become extremely difficult.
It is now possible to search for players by using (or excluding) diacritical characters in the search box. For example you can now search for Mané by either typing Mané or Mane in the search box.
An update on the rules changes is given on the new page.
21 out of the Top 50 managers have gone for Salah as captain, Fernandes 10, Martial 6, Rashford 5, Mané 3, Kane 1, Vardy 1, Alexander-Arnold 1, Mahrez 1, Pulisic 1
Another wild card onslaught is gaining momentum. Expect overly enthusiastic predictions whilst the volumes are high. Whenever the total volume of WC transfers for a player exceeds the effective transfers for that player, then predictions for that player become extremely difficult
10 out of the Top 50 managers have gone for Vardy as captain Abraham 7, Kane 6, Mané, Callum Wilson, Agüero, De Bruyne, Sterling 4, Jiménez 2, Alexander-Arnold, Haller, Mount, Hudson-Odoi, Pukki 1,
There will be a number of short outages today whilst I update and improve a number of programs that support the web site. The aim is to improve the accuracy of certain key parameters.
Next week I will be making some significant changes to the main pages on the site. The aim being to manage the excessive users loads after the matches complete. If all goes well no changes should be evident to the users, i.e each page will appear exactly as it did before.
Currently lots of wild cards (this is the norm at this stage of the season). So expect lots of overly enthusiastic predictions. Machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s.
Son finally rose. Wild Cards must have been the cause of the delayed change. A rule change is ruled out. The 12th Oct code changes have now been backed out.
Re: TAA last night - Prediction was 75.8%, error 24.2%. Currently at a loss to explain. Net transfers of ~-33 yesterday. So only a significant volume of WCs yesterday or this game week will explain. However, the machine learning components on the site falter at low net transfers.
Now hoping that Bruno, KDB, Martial et al do not rise this GW. This will explain some anomalies from last season when a number of players took ages to drop.
Murphy's FPL law means that the above will rise tonight.
Will make code changes tonight and run test cases at weekend.
We are now assuming that Son's reluctance to rise is due to a 'rule reveal'.
Code changes are in place. Expect Son's Tar% value (and a small number of other players) to jump around this afternoon.
If Son rises over the next couple of days we will back out the code changes.
All price change engines and machine learning algorithms that can be switched on (at this stage of the season) are now operational. The next suite will be switched on at the start of the next game week.
Found a few minor buglets. All are now fixed.
Inability to access to the site appears to be a DNS issue. If you CANNOT access the site use as normal use
http://217.194.223.2/
or just enter 217.194.223.2 in your browser
Kiko Femenía unexpected price rise is under investigation, but the reasons are not clear. No statistical errors are evident. Possible that a rule change is the cause, but no other players with a similar behaviour have been noted.
The total transfers for this game week are currently unexpectedly low (i.e. 35% down). Either teams are saving a transfer or have forgotten it's an early game start tonight.
There are currently a lot of transfers. Not decided yet if this due to wild cards. Just a warning that if it's due to wild cards then some of the predictions tonight may be a little optimistic.
@fpl_fail
Expected
1. Haaland is the same price as he started the season. If a player > start value he will fall quicker; or slower if <=.
2. The rate of speed is slowed further due to his level of ownership.
3. Flags also slow a player
4. Only effective transfers effect price changes.
9 out of the Top 50 managers have gone for Vardy as captain, Fernandes 8, Calvert-Lewin 8, De Bruyne 7, Salah 7, Kane 5, Rashford 1, Aubameyang 1, Sterling 1, Mané 1, Werner 1 and Foden 1
9 out of the Top 50 managers have gone for KDB as captain, Fernandes 8, Vardy 7, Calvert-Lewin 7, Kane 6, Salah 5, Jesus 3, Son 2, Werner 2 and Mahrez 1
The top 50 managers have played 76 chips (BB 22, FH 7, TC 25 and WC 22). Ten chips were played during the last game week. 19 of the top 50 have not played a chip (excl wild cards). Seven of the top 50 played their wild card during the int break.
To facilitate the analysis of the effectiveness of the transfers made (i.e. whether they contribute to the price changes) I need to produce a new programme. This will take three to four days. Got the design just need to do it.
Looks like the rule change mentioned on the 22nd Sept is real.
The main impacts are: i) players hovering close to the drop zone for extended periods but failing to drop in price; and ii) some rise errors.
It will be over a week before I can make the necessary adjustments.
There are lots of wild cards active (this is the norm early in the season). Expect lots of overly enthusiastic predictions. Machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s. Expect the biggest jumps to occur ~9 - 10pm GMT
16 out of the Top 50 managers have gone for Sterling as captain, Salah (9), Aguero (7), KDB (6), Kane (5), Hazard (2), Lukaku, de Gea, Firmino, Pogba, Sane (1)
#FPL
Implementing a lot of database changes and machine learning updates today. If the site fails to load it's because I'm updating the various system dlls. Each outage should last no more than 20 seconds.
Seem to have a bug with the systems this morning following the price changes. Going to have a quick look shortly. However, I'm not near my development machines, so it may be later this after afternoon before I can fully debug. .....but at least Ederson rose at last!!!
17 out of the Top 50 managers have gone for Sterling as captain, Aguero (14), Jimenez (9), KDB (3), Aubameyang, Son (2) and Lacazette, Mane, Eriksen (1)
A wave of wild cards is gaining momentum. It's not an onslaught, but enough to cause concern. Expect overly enthusiastic predictions whilst the volumes are high.
The top 50 managers have played 89 chips (BB 15, FH 11, TC 20 and WC 43). 3 chips were played during the last game week. Excluding wild cards only 19 of the top 50 teams have not played a chip. 5 have played all their chips.
There are currently lots of wild cards active (this is the norm early in the season). So expect lots of overly enthusiastic predictions. Machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s. Hopefully calmer later tonight.
14 out of the Top 50 managers have gone for Pulisic as captain, Sterling 7, Willian 7, Jesus 5, Fernandes 4, Foden 4, Martial 2, Alonso 2, Giroud 1, David Silva 1, De Bruyne 1, Salah 1, Ings 1
There are currently lots of wild cards active (this is the norm early in the season). So expect lots of overly enthusiastic predictions. Machine learning algorithms are making adjustments to compensate - hence the jumps in the prediction Tar%s. Hopefully calmer later today.
Only 16 of the Top 50 managers are serious players. 34 have played an excessive number of chips.
14 out of the Top 50 managers have gone for Sterling as captain, Agüero, Pukki 9, Salah 5, Son 3, Mané, Aubameyang, Mahrez 2 and Pépé, Haller, Abraham, Kane 1.
On a long haul flight tonight (Queensland time / morning UK time) for 36 hrs. Expect the prediction screens to get messy until I have internet and have overcome jet lag. Reduced AI in charge. All machine learning components will be decoupled shortly and reestablished in 48 hours.
The top 50 managers have played 59 chips (BB 11, FH 1, TC 46 and WC 1). Ten chips were played during the last game week. Only 1 of the top 50 teams have not played a chip. Clearly at this time the vast majority of the top 50 are not serious long turn title contenders.
@SpartacusFPL
@FPLHaul
The algorithms do change. When they do and it's confirmed by ourselves we announce a "rule change" or in some case a "rule reveal". I don't think the main site invokes "manual input" to make "life harder".
Yippee!!! - A test case has finally arrived (Lucas Moura) to help investigate the rule change observed on the 29th September. So if Lucas drops in price tonight you will know why! If he does drop in price then it's a strange rule.
Only 17 of the Top 50 managers are serious long term players. 33 have played an excessive number of chips.
18 (9 each) out of the Top 50 managers have gone for Kane or Abraham as captain KDB, Agüero 6, Mané 5, Vardy 4, C Wilson Sterling 3, Haller, Jiménez, Pukki, TAA, McGinn 1
Good Luck to everyone this season!!!
The systems are now in the calibration phase of learning. This phase will complete when both a price rise and a price drop have occurred
Good Luck to everyone for the 24/25 season!!!
The systems are now in the calibration phase of learning. This phase will complete when both a price rise and a price drop have occurred