Overall Net: A DNN for converting ratings to WS/48 and then Overall Grades (189 views)

[Deleted]

Overall WS48 Year Name
212 0.164 2012 Alexis Antetokounmpo
211 0.220 2011 Amare Stoudemire
207 0.227 2008 Champ Godbolt
200 0.211 2011 Cameron Reddish
198 0.190 2011 AJ Guyton
198 0.215 2012 Champ Godbolt
197 0.181 2010 Athansios Cathasach
192 0.205 2010 Champ Godbolt
191 0.228 2012 Cameron Reddish
189 0.232 2010 Brain Winter
188 0.260 2012 Al Harrington
188 0.211 2008 Amare Stoudemire
185 0.162 2011 Collin Sexton
185 0.216 2009 Brain Winter
184 0.192 2011 Brain Winter
184 0.127 2009 Alexis Antetokounmpo
184 0.158 2008 Athansios Cathasach
181 0.286 2012 Amare Stoudemire
181 0.215 2009 Champ Godbolt
181 0.106 2011 Alexis Antetokounmpo
180 0.199 2011 Al Harrington
179 0.239 2009 Amare Stoudemire
179 0.241 2010 Amare Stoudemire
177 0.203 2012 Athansios Cathasach
177 0.181 2010 Andre Turner
176 0.155 2010 AJ Guyton
175 0.179 2011 Champ Godbolt
174 0.148 2009 Cameron Reddish
171 0.179 2010 Cameron Reddish
170 0.181 2009 AJ Guyton
169 0.121 2009 Athansios Cathasach
169 0.202 2009 Collin Sexton
169 0.164 2011 Athansios Cathasach
169 0.128 2010 Alex Reese
168 0.171 2008 Brain Winter
167 0.136 2010 Caron Butler
165 0.129 2010 Alexis Antetokounmpo
164 0.127 2008 Alexis Antetokounmpo
164 0.179 2008 Cameron Reddish
164 0.154 2010 Al Harrington
158 0.141 2008 Caron Butler
157 0.192 2009 Al Harrington
156 0.190 2010 Collin Sexton
156 0.140 2008 AJ Guyton
155 0.208 2012 Caron Butler
154 0.120 2009 Antawn Jamison*
154 0.162 2008 Andre Turner
151 0.089 2009 Arvydas Sabonis
151 0.115 2009 Bison Dele
150 0.167 2011 Alex Reese
149 0.099 2009 Butch McRae
148 0.275 2012 AJ Guyton
148 0.150 2011 Anfernee Simons
148 0.149 2008 Cuttino Mobley
147 0.116 2011 Antawn Jamison
147 0.157 2008 Clarence Withers
146 0.146 2011 Bison Dele
144 0.143 2011 Andre Turner
144 0.195 2012 Andre Turner
142 0.120 2011 Benji Wilson
141 0.165 2009 Cody Zeller
140 0.077 2009 Alex Reese
140 0.122 2009 Cecil Exum
139 0.074 2011 Cecil Exum
139 0.138 2010 Antawn Jamison*
139 0.136 2011 Bol Bol
138 0.122 2008 Alex Reese
138 0.161 2012 Arvydas Sabonis
137 0.145 2008 Bol Bol
137 0.086 2010 Cecil Exum
136 0.054 2010 Benji Wilson
136 0.096 2011 Chris Porter
135 0.129 2009 Andre Turner
135 0.163 2008 Collin Sexton
133 0.161 2012 Alex Reese
133 0.108 2010 Chris Washburn
133 0.140 2010 Bison Dele
131 0.146 2010 Anfernee Simons
130 0.180 2008 Al Harrington
130 0.160 2009 Cameron Parker
129 0.086 2010 Chris Porter
128 0.113 2011 Arvydas Sabonis
128 0.115 2009 Caron Butler
125 0.134 2008 Antawn Jamison*
125 0.111 2008 Butch McRae
121 0.125 2009 Brandon Griner
121 0.134 2011 Cody Zeller
120 0.141 2011 Caron Butler
120 0.123 2008 Bobby McDermott
120 0.168 2012 Anfernee Simons
120 0.131 2009 Bol Bol
119 0.113 2010 Cuttino Mobley
117 0.070 2008 Charles Barkley*
117 0.114 2011 Bob Dandridge
115 0.147 2008 Brandon Griner
115 0.080 2011 Cuttino Mobley
114 0.154 2012 Chris Webber
114 0.092 2009 Cracker McCallahan
113 0.095 2010 Alex Schmidt
113 0.191 2012 Bison Dele
113 0.136 2010 Brandon Griner
112 0.088 2011 Chink Crossin
112 0.085 2008 Anthony Bennett
112 0.121 2009 Anfernee Simons
112 0.074 2008 Arvydas Sabonis
111 0.120 2009 Bobby McDermott
110 0.021 2008 Bison Dele
110 0.071 2008 Chris Chiozza
109 0.063 2008 Benji Wilson
108 0.102 2008 Anfernee Simons
108 0.123 2008 Cody Zeller
108 0.070 2008 Chris Porter
108 0.053 2009 Austin Wiley
106 0.079 2010 Cameron Parker
105 0.140 2011 Brandon Griner
105 0.101 2010 Bobby McDermott
103 0.112 2011 Chris Washburn
103 0.072 2009 Clarence Withers
101 0.115 2011 Ben McLemore
101 0.140 2008 Cracker McCallahan
101 0.051 2011 Ajdin Penava
100 0.103 2010 Bob Dandridge
98 0.094 2012 Byron Scott
98 0.070 2009 Chuck Person
97 0.102 2008 Chuck Person
96 0.108 2008 Cameron Parker
96 0.093 2010 Ajdin Penava
95 0.001 2010 Austin Wiley
93 0.082 2008 Byron Scott*
92 0.123 2010 Bol Bol
90 0.132 2012 Bobby Hurley
90 0.024 2008 Aljami Durham
90 0.102 2011 Alvin Jones
89 0.107 2009 Barry Obama
89 0.067 2009 Cuttino Mobley
87 0.093 2012 Bob Dandridge
87 0.095 2010 Clyde Drexler
86 0.101 2009 Benji Wilson
85 0.140 2008 Courtney Alexander
85 0.082 2008 Chandler Hutchison
85 0.054 2009 Charles Barkley*
85 0.022 2011 Carsen Edwards
83 0.112 2010 Courtney Alexander
80 0.068 2008 Bones McKinney
77 0.114 2010 Chink Crossin
77 0.093 2009 Courtney Alexander
74 0.064 2010 Ben McLemore
73 0.170 2011 Courtney Alexander
73 0.076 2011 Alex Schmidt
69 0.095 2009 Bob Dandridge
68 0.127 2008 Chink Crossin
67 0.006 2010 Carsen Edwards
67 0.102 2010 Alvin Jones
66 0.043 2010 Carlos Boozer
66 0.083 2010 Craig Raymond
65 0.111 2009 Byron Scott*
65 0.096 2011 Cliff Robinson
64 0.054 2011 Chandler Hutchison
63 0.073 2011 Byron Scott
63 0.063 2009 Bob Rule
61 0.048 2011 Clyde Drexler
61 0.038 2011 Ayo Dosunmu
58 0.140 2012 Alvin Jones
57 0.006 2009 Bud Ogden
57 0.037 2008 Ayo Dosunmu
56 0.082 2011 Carlos Boozer
56 -0.038 2010 Alvin Oakley
55 0.110 2010 Byron Scott*
54 0.044 2012 Chris Childs
52 0.106 2010 Ayo Dosunmu
51 -0.010 2009 Alvin Oakley
50 0.084 2008 Aiden Williamson
49 0.036 2011 Charles Barkley
48 0.016 2008 CJ McCollum
47 -0.035 2010 Carlos Arroyo
46 0.076 2008 Clyde Drexler
43 0.006 2010 Bud Ogden
41 0.083 2009 Aiden Williamson
41 0.020 2009 Ayo Dosunmu
41 0.010 2008 Ben McLemore
39 0.038 2011 Brian Scalabrine
39 0.044 2009 Brad Daugherty
36 0.036 2011 Brad Daugherty
35 0.099 2011 Aiden Williamson
33 0.007 2008 Carlos Boozer
33 0.047 2010 Chris Mihm
28 0.096 2010 Aiden Williamson
28 0.011 2011 Aljami Durham
26 -0.038 2009 Butch Beard
26 -0.037 2010 Butch Beard
25 0.056 2008 Carsen Edwards
24 0.006 2011 BJ Armstrong
23 -0.013 2011 Butch Beard
22 0.048 2009 Carlos Boozer
18 0.018 2011 Alfonso Wilkens
17 0.014 2009 CJ McCollum
13 0.031 2011 Brendan Haywood
8 0.016 2009 Bernie Williams
7 -0.016 2010 Brendan Haywood
2 0.052 2008 Anas Mahmoud
0 0.030 2011 Charlie Bell
0 0.075 2010 Brad Daugherty

docs.google.com/presentation/d/1C1_AdxLzxEth_FxsoDOSOu0QaQVv34R8Ax-AuXcTClo/edit?usp=sharing

[jhb]

25k

[eric]

alexis antetokounmpo jumps out as wildly overrated by the model, and was a good defender / middling scorer. can you look at error against predicted share of contribution due to defense-only attributes?

[Deleted]

Sept 13, 2019 17:40:39 GMT eric said:

alexis antetokounmpo jumps out as wildly overrated by the model, and was a good defender / middling scorer. can you look at error against predicted share of contribution due to defense-only attributes?

No you can't really do that. I can try redoing the model with less defensive attributes and quantifying the differences. I specifically remember @odin being really upset that Alexis was never as good as his ratings suggested, so I thought it was really cool to see that shown here.

[TimPig]

I wish I was smart. I feel like this would be a good article if I was.

25k

[Deleted]

This is nice, but a 0-100 scale is prettier and this doesnt include players who arent very good who would populate the low end of the spectrum

[eric]

Sept 13, 2019 17:41:54 GMT @bkisok said:

Sept 13, 2019 17:40:39 GMT eric said:

alexis antetokounmpo jumps out as wildly overrated by the model, and was a good defender / middling scorer. can you look at error against predicted share of contribution due to defense-only attributes?

No you can't really do that. I can try redoing the model with less defensive attributes and quantifying the differences. I specifically remember @odin being really upset that Alexis was never as good as his ratings suggested, so I thought it was really cool to see that shown here.

can you just send me the full data then? or just the formula and i'll apply it

[Deleted]

Sept 13, 2019 19:01:42 GMT eric said:

Sept 13, 2019 17:41:54 GMT @bkisok said:

No you can't really do that. I can try redoing the model with less defensive attributes and quantifying the differences. I specifically remember @odin being really upset that Alexis was never as good as his ratings suggested, so I thought it was really cool to see that shown here.

can you just send me the full data then? or just the formula and i'll apply it

That's not how it works. The formula is about 90 billion parameters and in the form of gigantic matrices

[Deleted]

Sept 13, 2019 18:49:13 GMT @odin said:

This is nice, but a 0-100 scale is prettier and this doesnt include players who arent very good who would populate the low end of the spectrum

You could easily scale this from 0->100

Your point about not including low end players is notable however, as 0 is the grade for a player that gets at least 1000 minutes while getting near negative winshares.

[TimPig]

Sept 13, 2019 19:32:56 GMT @bkisok said:

1000 minutes while getting near negative winshares.

Lmao Brad Daugherty

[eric]

Sept 13, 2019 19:32:33 GMT @bkisok said:

Sept 13, 2019 19:01:42 GMT eric said:

can you just send me the full data then? or just the formula and i'll apply it

That's not how it works. The formula is about 90 billion parameters and in the form of gigantic matrices

so send me that, whatever takes the inputs and produces an output

[Deleted]

Sept 13, 2019 20:11:00 GMT eric said:

Sept 13, 2019 19:32:33 GMT @bkisok said:

That's not how it works. The formula is about 90 billion parameters and in the form of gigantic matrices

so send me that, whatever takes the inputs and produces an output

You have the link to that it's on my github. You're gonna need to install a lot of dependencies to run it though. I could send you a primer. Clone the github repo, the training/validation are in data folder. You just need to run the python script. I suggest install anaconda3 and use conemu as your terminal. Then just open the terminal in the folder where you downloaded the script and run it. It'll give you errors like "you need to install tensorflow" then type the command to install that inside of your anaconda virtual enviroment, then run it again. It'll give you another error like "you need to install ____". Install it. Repeat until it runs. Then it'll output a csv called "foo.csv" and that will have the outputs.

Tldr:
Anaconda3
Conemu
Conda install pip
Pip install (all error dependencies)
Python overall net

[TimPig]

[Deleted]

Sept 13, 2019 21:58:12 GMT @bkisok said:

Sept 13, 2019 20:11:00 GMT eric said:

so send me that, whatever takes the inputs and produces an output

You have the link to that it's on my github. You're gonna need to install a lot of dependencies to run it though. I could send you a primer. Clone the github repo, the training/validation are in data folder. You just need to run the python script. I suggest install anaconda3 and use conemu as your terminal. Then just open the terminal in the folder where you downloaded the script and run it. It'll give you errors like "you need to install tensorflow" then type the command to install that inside of your anaconda virtual enviroment, then run it again. It'll give you another error like "you need to install ____". Install it. Repeat until it runs. Then it'll output a csv called "foo.csv" and that will have the outputs.

Tldr:
Anaconda3
Conemu
Conda install pip
Pip install (all error dependencies)
Python overall net

Actually I remember you were using python for some other stuff? However you are running that would probably work. I hope it was with anaconda. If so, just pip install the dependences that are imported at the beginning of the python script. You'll have to train it yourself for now unless you want me to send you the model but it won't run unless you can train it yourself so you might as well get it to that point. Once you do you can choose to "eval" instead of "train" and itll just read whatever players and ratings are in the "validation.csv" and output too foo.csv. However you can simply change the

dependencies you need:

import tensorflow as tf # tensorflow module
import numpy as np # numpy module
import pandas as pd
import math
import os # path join

change this line to have the eval function read from a different csv:
val_data = pd.read_csv('./data/Validation.csv')

just comment out the onet.train line to have it run just the evaluation:

Onet.train()   ---> #Onet.train()
Onet.eval()