CORRELATION

[Denwen12]

i have a correlation file i build, ill give an example Anyway i can get this optimizer to use this. if a player has a minus correlation i don't want them in the same lineup. Is this a possiblity

Player	MP.Ply	Teammate	MP.Tmm	Team	corMin	corDK	corFD
Trae Young	1981	De'Andre Hunter	1799	ATL	0.421	0.124	0.087
Trae Young	1854	Vince Carter	758	ATL	-0.181	-0.149	-0.114
Trae Young	1829	Cam Reddish	1384	ATL	0.022	0.024	0.04
Trae Young	1813	Bruno Fernando	649	ATL	-0.264	-0.031	-0.053
De'Andre Hunter	1799	Trae Young	1981	ATL	0.421	0.124	0.087
De'Andre Hunter	1795	Vince Carter	811	ATL	-0.049	0.151	0.128
De'Andre Hunter	1746	Cam Reddish	1467	ATL	-0.141	0.078	0.015
De'Andre Hunter	1745	Bruno Fernando	700	ATL	-0.072	-0.029	-0.061
Trae Young	1733	Kevin Huerter	1529	ATL	0.296	0.191	0.189

[cdbmgrdn]

https://pydfs-lineup-optimizer.readthedocs.io/en/latest/rules.html#group-players

It looks possible to do it by this method. Get the players you can't have in the same lineup from your correlation sheet and then it should be possible to loop through that method with your list of 2 players.

[Denwen12]

im not sure its possible cause my file is pretty big

[cdbmgrdn]

It should be fine but the optimizer might take longer. You'll just have to iterate through that list and create player groups before the optimizer generates the lineups. Should be pretty easy to do with Pandas. Maybe try adding this code:

import pandas as pd

csv_filepath = 'your correlation csv'

cdf = pd.read_csv(csv_filepath)

#set this to whatever you want the correlation cutoff to be
correlation_threshold = -0.15

for i in range(len(cdf)):
    if cdf['corDK'].iloc[i] <= correlation_threshold:
        negative_corr_players = (cdf['Player'].iloc[i],cdf['Teammate'].iloc[i])
        group = PlayersGroup([optimizer.get_player_by_name(name) for name in (negative_corr_players)])
        optimizer.add_players_group(group)

[Denwen12]

thank you, ill try this out after work today

[Denwen12]

I been working on this and i have to set the threshold like -.800 to get it to make a lineup so seems like its not working and i also deleted the duplicate lines , if i lower the threshold to where i want it i get

root@localhost:~/NBA/2Slate# python3 Fanduel2Finish.py Traceback (most recent call last): File "Fanduel2Finish.py", line 97, in lineups = list(optimizer.optimize(n=1,)) File "/usr/local/lib/python3.7/dist-packages/pydfs_lineup_optimizer/lineup_optimizer.py", line 395, in optimize constraint.apply(base_solver) File "/usr/local/lib/python3.7/dist-packages/pydfs_lineup_optimizer/rules.py", line 302, in apply self._create_constraints(solver) File "/usr/local/lib/python3.7/dist-packages/pydfs_lineup_optimizer/rules.py", line 336, in _create_constraints variables = [self.players_dict[p] for p in group_players] File "/usr/local/lib/python3.7/dist-packages/pydfs_lineup_optimizer/rules.py", line 336, in variables = [self.players_dict[p] for p in group_players] KeyError: None

[cdbmgrdn]

How big is the slate you're trying to create lineups for? There might not be possible combinations available on smaller slates which do not have negatively correlated players.

[Denwen12]

it was todays slate with 7 games so it should be possible but its not i gotta set -.800 to get it to work

On Thu, Dec 17, 2020, 5:07 PM cdbmgrdn notifications@github.com wrote:

How big is the slate you're trying to create lineups for? There might not be possible combinations available on smaller slates which do not have negatively correlated players.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-747731102, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7XMKFJBTIKABM3AGA3SVJ6KRANCNFSM4UY5MIBQ .

[Denwen12]

Possibly it can be the names not matching

On Thu, Dec 17, 2020, 6:47 PM joseph offen joffen12@gmail.com wrote:

it was todays slate with 7 games so it should be possible but its not i gotta set -.800 to get it to work

On Thu, Dec 17, 2020, 5:07 PM cdbmgrdn notifications@github.com wrote:

How big is the slate you're trying to create lineups for? There might not be possible combinations available on smaller slates which do not have negatively correlated players.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-747731102, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7XMKFJBTIKABM3AGA3SVJ6KRANCNFSM4UY5MIBQ .

[cdbmgrdn]

I'm working on something similar and I'm now thinking the issue could be with the types (I ran into a very similar problem).

Make sure that each list of uncorrelated players turns into a PlayersGroup. optimizer.add_players_group() needs a PlayersGroup input and not a list.

I'm thinking that setting your correlation cutoff to -.800 just shrunk your list to zero which made it work.

Try something like this if you have a list of lists fo uncorrelated players:

for i in range(len(uncorrelated_players_list)):
    uncorrelated_players_list[i] = PlayersGroup(uncorrelated_players_list[i])

[Denwen12]

i can upload my list if you want

On Fri, Dec 18, 2020, 1:56 PM cdbmgrdn notifications@github.com wrote:

I'm working on something similar and I'm now thinking the issue could be with the types (I ran into a very similar problem).

Make sure that each list of uncorrelated players turns into a PlayersGroup. optimizer.add_players_group() needs a PlayersGroup input and not a list.

I'm thinking that setting your correlation cutoff to -.800 just shrunk your list to zero which made it work.

Try something like this if you have a list of lists fo uncorrelated players:

for i in range(len(uncorrelated_players_list)): uncorrelated_players_list[i] = PlayersGroup(uncorrelated_players_list[i])

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-748261440, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7SYHXNR7MQCPS5QKFTSVOQWFANCNFSM4UY5MIBQ .

[Denwen12]

whats your email address? that's much better for me to communicate

On Fri, Dec 18, 2020, 3:00 PM joseph offen joffen12@gmail.com wrote:

i can upload my list if you want

On Fri, Dec 18, 2020, 1:56 PM cdbmgrdn notifications@github.com wrote:

I'm working on something similar and I'm now thinking the issue could be with the types (I ran into a very similar problem).

Make sure that each list of uncorrelated players turns into a PlayersGroup. optimizer.add_players_group() needs a PlayersGroup input and not a list.

I'm thinking that setting your correlation cutoff to -.800 just shrunk your list to zero which made it work.

Try something like this if you have a list of lists fo uncorrelated players:

for i in range(len(uncorrelated_players_list)): uncorrelated_players_list[i] = PlayersGroup(uncorrelated_players_list[i])

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-748261440, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7SYHXNR7MQCPS5QKFTSVOQWFANCNFSM4UY5MIBQ .

[Denwen12]

i could also be looking at the numbers incorrectly. Is a negative better or the postive ?

On Fri, Dec 18, 2020, 4:56 PM joseph offen joffen12@gmail.com wrote:

whats your email address? that's much better for me to communicate

On Fri, Dec 18, 2020, 3:00 PM joseph offen joffen12@gmail.com wrote:

i can upload my list if you want

On Fri, Dec 18, 2020, 1:56 PM cdbmgrdn notifications@github.com wrote:

I'm working on something similar and I'm now thinking the issue could be with the types (I ran into a very similar problem).

Make sure that each list of uncorrelated players turns into a PlayersGroup. optimizer.add_players_group() needs a PlayersGroup input and not a list.

I'm thinking that setting your correlation cutoff to -.800 just shrunk your list to zero which made it work.

Try something like this if you have a list of lists fo uncorrelated players:

for i in range(len(uncorrelated_players_list)): uncorrelated_players_list[i] = PlayersGroup(uncorrelated_players_list[i])

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-748261440, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7SYHXNR7MQCPS5QKFTSVOQWFANCNFSM4UY5MIBQ .

[Denwen12]

What is Teammate Correlation?

The Teammmate Correlation tool looks at the relationship between two teammates’ performances (minutes per game, DraftKings points per game and FanDuel points per game), measured by correlation.

Correlation measures how two values move relative to each other, and is expressed as the correlation coefficient, a number on a scale from -1 (perfectly negative correlation) to 1 (perfectly positive correlation).

The closer the correlation coefficient is to 0, the weaker the relationship, and the further it is from 0, the stronger the relationship.

For a positive correlation, the values move in the same direction. If Value A increases, Value B increases. If Value A decreases, Value B decreases.

For a negative correlation, the values move in opposite directions. If Value A increases, Value B decreases. If Value A decreases, Value B increases.

Finally, there are also cases of no correlation. In these cases, a change in Value A doesn’t impact Value B.

A final point on correlation: we are not dealing with any perfectly correlated relationships. For example, in the positive correlations we will look at below, when Value A moves one direction, Value B will typically move in the same direction, but not always.

The closer the correlation coefficient is to 1 (or -1), the more often the relationship will be true.

The general idea is: how does a change in Player A’s performance impact our expectation for Player B’s performance? If we expect Player A to have a big game tonight (which we do if we’re playing them in a large field tournament), how does that impact our prediction for Player B?

On Fri, Dec 18, 2020, 6:06 PM joseph offen joffen12@gmail.com wrote:

i could also be looking at the numbers incorrectly. Is a negative better or the postive ?

On Fri, Dec 18, 2020, 4:56 PM joseph offen joffen12@gmail.com wrote:

whats your email address? that's much better for me to communicate

On Fri, Dec 18, 2020, 3:00 PM joseph offen joffen12@gmail.com wrote:

i can upload my list if you want

On Fri, Dec 18, 2020, 1:56 PM cdbmgrdn notifications@github.com wrote:

I'm working on something similar and I'm now thinking the issue could be with the types (I ran into a very similar problem).

Make sure that each list of uncorrelated players turns into a PlayersGroup. optimizer.add_players_group() needs a PlayersGroup input and not a list.

I'm thinking that setting your correlation cutoff to -.800 just shrunk your list to zero which made it work.

Try something like this if you have a list of lists fo uncorrelated players:

for i in range(len(uncorrelated_players_list)): uncorrelated_players_list[i] = PlayersGroup(uncorrelated_players_list[i])

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DimaKudosh/pydfs-lineup-optimizer/issues/229#issuecomment-748261440, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBWO7SYHXNR7MQCPS5QKFTSVOQWFANCNFSM4UY5MIBQ .

[Denwen12]

what is the full code i should try

[sansbacon]

A couple of comments, one theoretical and one practical.

The theoretical comment is that the "no negatively correlated players" rule is likely a bad idea, especially as the slate size goes down. Most correlation coefficients for teammates are statistically weak, they do not measure the magnitude of the relationship, and do not take into account salary or the other players on the team. You may not want to play $8,000 players together if they have negative correlation, but it may be a mistake to pass on a $3,500 player on this basis.

The practical comment is that these types of relationships are better expressed in projections rather than optimizer rules. The optimizer rule "no negative correlations" can be approximated by generating projections that downgrade player B when player A has a favorable projection, while also allowing these players to be played together when justified by salary and the adjusted projections. This approach should also be superior from a performance standpoint, as creating more and more complicate optimizer rules can slow things down substantially.