Jump to content
Sports Interactive Community

mike_e

Members
  • Content Count

    15
  • Joined

  • Last visited

About mike_e

  • Rank
    Amateur
  1. I've been experiencing the same problem as the OP, and this was a really helpful. With one difference....it wasn't players being scouted, but rather players being analysed. When I went to Scouting | Assignments | Players, I found I had over a hundred players being analysed. I removed them all, and matchday processing has gone from minutes to seconds. Not sure why there were so many players being analysed, though. It looks like they accumulated over time. Perhaps there should be logic to clear them out after a short period of time?
  2. For previous versions, I'd read that the player needs to play long enough to get a rating in the match in order for the game time to "count" towards development. Is that still the case?
  3. Unfortunately, none of the media-handling types give any indication about ambition. Which is a shame.
  4. Well, pulling the data proved to be more tedious than I'd expected, due to attribute masking. To simplify, I changed how I collected the data. Unfortunately, this introduced differences between the this data set and the previous one. These differences were quite big. For example, the model that explained 50% of the variance in the first data set only explains 30% of the variance in the second data set. I'd planned to go back and really think through how to collect the data so as to eliminate collection biases. But then the FM16 Beta arrived, and I got... distracted. Actually, I doubt I'll go back now and re-do the data collection. We're stuck with this data, despite the dubious quality. The theory we want to test is whether there are attributes that contribute to a player's development by affecting the amount of playing time he gets. Running a regression against all the attributes (visible and hidden) gives a model that explains 50-60% of the variance. I've pasted the full results below. As before, we see Injury Proneness is the biggest factor. Second is Stamina. Both of these seem like they are contributing to the player's development by affecting playing time. Natural Fitness is also a factor, which is another attribute that should affect playing time. So, this seems to supports HUNT3R's theory that there are attributes that appear to contribute to player development, but in fact only do so indirectly, by affecting playing time. And it interesting to note that Det is no longer a large contributing factor. It *may* be a proxy for playing time, but not to the degree that Injury Proneness, Stamina, and Natural Fitness are. Seems to me we really need playing time data as part of this analysis. Which is really hard to get in an easily automated fashion. Here's the data in full... Estimate Std. Error t value Pr(>|t|) (Intercept) -6.216e-02 1.545e-01 -0.402 0.68800 Prof 7.744e-03 3.486e-03 2.221 0.02784 * Amb 3.267e-03 3.407e-03 0.959 0.33918 Inj.Pr -1.635e-02 2.990e-03 -5.468 1.89e-07 *** Cor 1.125e-02 4.611e-03 2.439 0.01590 * Cro 2.848e-03 3.589e-03 0.794 0.42874 Dri -7.161e-03 4.555e-03 -1.572 0.11808 Fin -1.973e-03 4.271e-03 -0.462 0.64476 Fir -9.463e-03 5.990e-03 -1.580 0.11629 Fre 2.621e-03 4.356e-03 0.602 0.54835 Hea -4.877e-04 3.393e-03 -0.144 0.88589 Lon 4.271e-03 4.027e-03 1.061 0.29063 L.Th -9.762e-04 3.665e-03 -0.266 0.79031 Mar -1.738e-03 4.832e-03 -0.360 0.71960 Pas 2.526e-03 4.407e-03 0.573 0.56742 Pen 1.559e-03 3.850e-03 0.405 0.68610 Tac 6.974e-03 5.493e-03 1.270 0.20621 Agg -8.365e-04 2.560e-03 -0.327 0.74431 Ant 1.011e-03 4.222e-03 0.240 0.81104 Bra 1.331e-04 2.270e-03 0.059 0.95332 Cmp 9.614e-03 4.540e-03 2.117 0.03590 * Cnt -2.486e-03 4.468e-03 -0.556 0.57873 Dec 3.370e-04 5.165e-03 0.065 0.94807 Det -1.628e-03 3.385e-03 -0.481 0.63118 Fla 5.351e-03 3.626e-03 1.476 0.14220 Ldr 4.586e-05 2.799e-03 0.016 0.98695 OtB 9.063e-03 4.709e-03 1.925 0.05620 . Pos 7.134e-03 5.885e-03 1.212 0.22737 Tea -3.161e-03 3.586e-03 -0.882 0.37939 Vis -1.404e-02 4.879e-03 -2.879 0.00459 ** Acc 1.161e-02 4.750e-03 2.444 0.01569 * Agi 1.835e-03 6.057e-03 0.303 0.76241 Bal 4.905e-03 4.346e-03 1.129 0.26087 Jum -3.353e-04 3.639e-03 -0.092 0.92672 Nat 9.385e-03 4.001e-03 2.346 0.02033 * Pac 3.912e-03 6.473e-03 0.604 0.54654 Sta 1.802e-02 5.152e-03 3.498 0.00062 *** Str 9.417e-03 5.968e-03 1.578 0.11675 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.1153 on 148 degrees of freedom Multiple R-squared: 0.609, Adjusted R-squared: 0.5113 F-statistic: 6.231 on 37 and 148 DF, p-value: 3.447e-16
  5. I first identified the types of players that I wanted for my tactical system (I had seven of these types), and then set up weights for each type. I then tweaked the weights over a number of iterations. I could definitely see the effect of changing the weights over time---the final set of weights gave "better" ratings than the first set ("better" being entirely subjective, of course). Yes. If the roles had the same weights, he combined them. These are the mean and standard deviation of "historical" data. They are used to normalize ratings between different roles. In the comments of his post, he says that "they’re from several of my teams. I could do all the top European leagues but it’d take ages". I did something similar, although I don't recall exactly what set of teams I used. Not necessarily. As I mentioned above, I came up with my own weights entirely. You can completely customize to your own judgement and preferences. It definitely helped me do that with a number of players.
  6. Here's a post about using a weighted rating system to evaluate players: https://fmcoffeehouse.wordpress.com/2014/12/14/fm15-1718-dortmund-always-the-bridesmaid-never-the-bride/#more-371. Look for the section titled 'The Rating System: How does it work'. I used an approach based on this post for my long-term FM15 save. I used it to analyze my own players, and also for scouting. It worked pretty well. As the post points out, you can't blindly rely on the numbers. You still have to use your own judgement.
  7. Interesting. It's relatively easy to extract the data to test that theory. I'll take a stab at it.
  8. Unfortunately, my browser glitched while I was typing up my analysis, and I lost about half of it. When I re-wrote it, I left out a really important caveat. So here it is... The R value of the regression model is about 50%. That means that this model "explains" about 50% of the variability in the increase in CA. The other 50% is due to factors that are not modeled, such as playing time, quality of opposition, quality of coaching, and other factors. Looking at it another way, this model describes the affect of Prof, Amb, Det, and Injury Proneness on development, assuming that all other factors are constant (on average). I'm definitely not suggesting that when we play the game we should ignore these other factors. They are crucially important to development. I agree, with one distinction. Based on the analysis, I would say that determination has no *direct* effect on development, but it does have an indirect effect. The Det*Inj.Pr term suggests that Det can modify the impact of Injury Proneness. And Injury Proneness, as you point out, has a clear indirect effect on development. It is possible that both Det and Det*Inj.Pr are acting as a proxy variables for playing time, and that their significance would be reduced if playing time were added to the model. Unfortunately, extracting playing time data from the game is a bit tedious, which is why I omitted it from the model in the first place.
  9. A while back I pulled some regen data, to look at exactly this issue. This is data from a long-term save, looking at the progress of three years of regens over a period of 10 years. All regens (total of 168) started their careers at a Premier League club, and after the 10 years were at a club somewhere in the top six leagues in England. The regens were from 2015, 2016, and 2017. They were evaluated in 2025, 2026, and 2027 respectively. For each regen, I have their Det, all hidden attributes, starting CA, ending CA, and PA. Take a look, the spread-sheet is here: https://docs.google.com/spreadsheets/d/14Rva5yIJWaXGqKZpq_V2r21VEPyEb6aYHXX7HX-Payc/edit?usp=sharing I'm not a statistician, but I have a copy of R and some time on my hands, so I've taken a stab at doing some analysis. I've calculated how much each regen increased their CA, scaled by how much they *could* have increased. Then I ran a multivariate regression on the hidden attributes plus Det against the percentage increase. Here's what R spat out: Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.2919795 0.1424770 -2.049 0.042140 * Det 0.0168426 0.0042602 3.953 0.000117 *** Prof 0.0360109 0.0051025 7.057 5.53e-11 *** Amb 0.0221242 0.0045349 4.879 2.66e-06 *** Cont 0.0019950 0.0049314 0.405 0.686372 Ada 0.0059487 0.0043033 1.382 0.168879 Cons 0.0050925 0.0050956 0.999 0.319186 Dirt 0.0004178 0.0043732 0.096 0.924006 Imp.M -0.0007775 0.0039183 -0.198 0.842971 Inj.Pr -0.0291618 0.0040486 -7.203 2.50e-11 *** Loy 0.0052854 0.0044698 1.182 0.238858 Pres -0.0007325 0.0054872 -0.133 0.893976 Spor -0.0052793 0.0045982 -1.148 0.252715 Temp 0.0066179 0.0049603 1.334 0.184131 Vers 0.0054448 0.0054712 0.995 0.321217 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.1778 on 153 degrees of freedom Multiple R-squared: 0.542, Adjusted R-squared: 0.5 F-statistic: 12.93 on 14 and 153 DF, p-value: < 2.2e-16 The attributes flagged with *** are the significant ones, all the others are irrelevant. So Prof, Injury Proneness, Amb, and Det all have an effect on player development (in that order of importance). We're looking for high Prof, Amb, and Det, and low Inj Proneness. Prof, Amb, and Injury Proneness work exactly as expected, but Det is a surprise. Based on earlier posts in this thread, and everything else I've read, I would not expect Det to have an effect. Looking for an explanation, I noticed that there is a second order effect between Det and Injury Proneness. If we add this to the model (and get rid of all the un-interesting attributes), we get the following: Residuals: Min 1Q Median 3Q Max -0.43410 -0.11319 0.00871 0.11658 0.47862 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.250665 0.160709 1.560 0.12077 Det -0.006877 0.009671 -0.711 0.47805 Prof 0.036871 0.004676 7.886 4.36e-13 *** Amb 0.022405 0.004328 5.177 6.60e-07 *** Inj.Pr -0.068780 0.013863 -4.961 1.75e-06 *** Det:Inj.Pr 0.002979 0.001009 2.952 0.00363 ** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.1736 on 162 degrees of freedom Multiple R-squared: 0.5374, Adjusted R-squared: 0.5232 F-statistic: 37.64 on 5 and 162 DF, p-value: < 2.2e-16 What this tells us is that Injury Proneness, Prof, and Amb are still significant, but Det is no longer considered significant. Rather, the interaction between Det and Injury Proneness (labeled Det:Inj.Pr) is significant. What this *might* mean is that Det mitigates the effect of Injury Proneness. In game terms, perhaps a high Det modifies the probability that an injury will occur, representing the willingness of a player to "play through the pain". This would mean that a high Det regen would gain valuable playing time compared to his low Det counterpart. So, if you want to improve a your regens chances of reaching their full potential, this data suggests doing the following: Prioritize high Professionalism; Focus on regens with low Injury Proneness; Prioritize high Ambition, but not at the expense of high Professionalism; If you have an Injury Prone regen, prioritize high Determination.
  10. Here's a bit of a short cut...control-click on all you players as you have been doing. Then, in the *next* match, if you control-click on the ball, the entire team will have their names displayed. This is useful, but not as useful as it could be. There's no way to guarantee *which* team will get their names displayed. Could be you, could be the other team. It seems pretty random. Still, it works in half the games. Another problem is that you can only do it once per match. If you lose the names for whatever reason, then you have to add them back one by one. Not great, but mildly useful.
  11. I was also a little disappointed with my youth intake in my last save, so I pulled some data for two years of youth intakes for EPL teams into a spreadsheet. I've made it public here: https://docs.google.com/spreadsheets/d/1Jib73uhBOmYufpbZwl3-kSDFIURvrl-5ErUjoWeplhQ/edit?usp=sharing The short version is that very few teams get good regens. In the two years that I looked at, there were only two regens over 160, and fourteen over 150. Chelsea has by far the best intakes: half of the 150+ regens went to Chelsea. Interestingly, they are the only EPL team with a 5* reputation, which perhaps guarantees them the bulk of the good regens from London?
  12. As one of the people arguing for using attributes over statistics, I definitely misunderstood what you were attempting. I thought that you were claiming that using statistics was the optimal way to play. I now see that you're attempting to follow a real world approach as a self-imposed constraint, regardless of whether or not it is optimal. I think that's really interesting, and I'm keen to hear how things progress. One thought....have you look at what the CIES Football Observatory is doing? They have a system of rating players, plus a financial model for predicting transfer costs. There's not a ton of detail available, but at a high level they rate players in six categories, and then take different weightings of the categories based on positions. They define the categories as follows: Shooting: ability to exploit goal opportunities through accurate shooting. Chance creation: ability to putting teammates in a favourable position to score. Take on: ability to create advantageous situations by successfully challenging opponents. Distribution: ability to keep a hold on the game through efficient passing. Recovery: ability to minimise opponents’ chances through proficient interception work. Rigour: ability to minimise opponents’ chances through robust duelling. Looking back at the list of metrics that you mentioned in your OP, these line up quite well: Shooting: Non-Penalty goals, Shots, Shooting%, Goal Conversion% Chance creation: Assists, Key passes Take on: Successful Dribbles Distribution: Passing% Recovery: Int Rigour: tackles The site is: http://www.football-observatory.com. The document describing what they are doing is here: http://www.football-observatory.com/IMG/pdf/cies_footobs_eng.pdf
  13. I've used a sort of money-ball approach in my current save. Like the OP, I started by using stats to rate each player. I couldn't find a way to make it work, though. You want to be able to use the rating to compare players on different teams, in different positions/roles/duties, different formations, different leagues, different levels, etc. Perhaps it is possible to control for all those variables, but I couldn't even start to figure out how. So I came to the same conclusion as sebs: use attributes, not stats, to rate players. I use a weighted average of attributes, with different weights for each position/role/duty that I use in my formation. It would be hard to come up with the "perfect" set of weights, but a "good enough" set of weights is easy enough to arrive at. Once you have the rating system in place, you can definitely find good players who are undervalued by the game. With a little help from the "Print to web page" option, you can export masses of player data, and use the rating to hone in on the best players in your price/wage range. I've used this approach for 5 or 6 seasons now, with decent results. Interestingly, while the stats-based approach will lead you to players coming off a good season, the attributes-based approach will lead you to players who've had a mediocre season (since their price will be depressed). That feels like money-ball, at least as I understand it. Mike
  14. Hi Wellingman. Sorry for the slow response. Did you make much progress with your save? Did you manage to stay up? I've had some good results in my game. Early in my third season in the Conference my 4-5-1 "clicked". I shot up the table (from 16th place in November), finished 2nd, and won promotion in the play-offs. This despite a media prediction of 14th. Now I'm mid-way through my first season in League 2, and I'm (unbelievably) in first place after 20 games. This time the media prediction is 24th, so it is even more surprising. Admittedly, I've won a lot of close games, so I've been a bit lucky. Sadly, I can't claim any great tactical revelations. The line-up is similar to the one I posted earlier, except that the wingers are wide midfielders (suits the players better), and the CM(S) is replaced by a DLP(S). The idea of using two DLPs came from Jambo98's excellent 4-5-1 thread, and seems to have solved the problem of defenders playing aimless long balls. I've experimented a lot with tempo, defensive line, and closing down. I like the balance that I found, using either Standard or Control, adjusted by TIs and PIs. It's a simple setup, but it seems pretty solid (watch me go on a losing streak now). You asked about the wb(a) role. I've had no problems with it. The deepest lying midfielder is on that side, and so offers some cover. Are you still using the LFB? I've never used one. They seem too....limited. Mike
  15. Hi Wellingman, I'm in roughly the same situation. Got promoted from the VCS with Boreham Wood. The first six months in the VC were actually surprisingly easy. I played a 4-1-4-1 attacking formation, and was in the playoff positions in December. Then it all went wrong, and I finished in 12th. This season I switched to a 4-5-1 control/attacking formation, and it is tough going. I'm just above relegation. It might just be a co-incidence, but I started struggling after moving to the 15.1.4 hotfix. Here's the set-up that I'm using: DLF(A) W(A) DLP(D) CM(A) CM(S) W(A) WB(A) CD(D) CD(D) FB(S) GK(D) The GK and back line are set to pass shorter. The DLF is set to roam. Usually control, but I'll switch to Attacking or Counter depending on the needs of the game. Sometimes it works well, but it is inconsistent. The players, except for the DLF, defend really deep. When we win possession, there are no outlets, and an aimless long ball follows. The obvious fix, playing a higher line, doesn't work, because my defenders are not fast enough to handle balls over the top. So I'm experimenting with all the options that force the team to play out of the back. I think I'm seeing some improvement, but there's still a long way to go. My financial experience has been completely the opposite of yours. The club is making money hand over fist. Halfway through my third season, and the club is valued at 1.5 million. I'm not sure what it was when I started, but I think it was under 100k. I've yet to buy a player, building the squad through frees, out-of-contract, and loan players. I force myself to stay well under the wage budget; right now it is 440k per year, and I'm spending about 380k. Not sure if that's of any help, but I thought I'd commiserate, at least.
×
×
  • Create New...