LOL win prediction
0 likes152 views
League of legend win prediction on master 1 tier. A pipeline of data analytics and how to improve results.
![襷 螻褐 == 一危 り
{
"matchDuration": 1716, // 豐 , 朱 蟆 る 讌螳襯 覩
"teams": [ // 2螳 Dictionary
{
"firstDragon": true, // (覈ろ)襯 襾殊 譯曙螳
"dragonKills": 2, // 覈覯 譯曙螳
"winner": false, // 豪Μ螳
"firstBaron": false, // 覦襦(覈ろ)襯 襾殊 譯曙螳
"baronKills": 0, // 覦襦 覈覯 譯曙螳
"firstBlood": false, // 豌 螳
"teamId": 100 // ID
},
...
],
"participants": [ // 10螳 Dictionary
{
"championId": 412, // 豈殊 (貂襴) ID
"summonerId": 21983, // 危 蟆企┯ ID
"teamId": 100, // ID (teams teamId 襷れ)
"stats": { // 企 糾襯 覩
"kills": 2, // 蟆企┯螳 るジ 蟆企┯襯 覈覯 譯曙螳
"deaths": 8, // 蟆企┯螳 覈覯 譯曙螳
"assists": 11, // るジ 蟆企┯襯 譯曙企 覈覯 譯殊螳
"goldEarned": 7314, // 豐 螻 (蟆 )
"totalDamageDealt": 27629, // 覈ろ + 蟆企┯蟆 螳 豐 磯語
"totalDamageDealtToChampions": 9507, // 蟆企┯蟆襷 螳 豐 磯語
"totalDamageTaken": 20419, // 覦 豐 狩
"minionsKilled": 37, // 譯曙 覩碁(覈ろ)
"totalHeal": 1014, // 豐 豺
"totalTimeCrowdControlDealt": 241, // るジ 伎伎蟆 Crowd Control (CC) 襯 豐 螳
"wardsPlaced": 5 // Ward(朱ゼ 譯朱 危) 襷旧 レ逢
"items": [ // 危 覲
3401,
2049,
1031,
3270,
0,
2043,
{
"teams": [ // 2螳 給.
{
"teamId": 200 // ID
},
...
],
"participants": [ // 10覈 蟆企┯ 覲
{
"championId": 421, // 豈殊 ( 貂襴) ID
"summonerId": 22082, // 蟆企┯ ID
"teamId": 100 // ID
},
...
]
}](https://image.slidesharecdn.com/random-180115015100/85/LOL-win-prediction-5-320.jpg)
![Experiment. Iter 0
{
"teams": [ // 2螳 給.
{
"teamId": 200 // ID
},
...
],
"participants": [ // 10覈 蟆企┯ 覲
{
"championId": 421, // 豈殊 ( 貂襴) ID
"summonerId": 22082, // 蟆企┯ ID
"teamId": 100 // ID
},
...
]
}
Test 覦 覲企
豈殊 ID, 蟆企┯ ID,
<螳 襴>
蟆 磯襯企ゼ 覦危讌 螻
譴 蟇 豈殊語願讌?
1 0 0 1 0
豈殊
0 0 1 1 0
豈殊
+
Team 100 Team 200
Sparse Matrix](https://image.slidesharecdn.com/random-180115015100/85/LOL-win-prediction-8-320.jpg)
![Experiment. Iter 1
{
"teams": [ // 2螳 給.
{
"teamId": 200 // ID
},
...
],
"participants": [ // 10覈 蟆企┯ 覲
{
"championId": 421, // 豈殊 ( 貂襴) ID
"summonerId": 22082, // 蟆企┯ ID
"teamId": 100 // ID
},
...
]
}
豈殊 覲, 覲企 譴朱
蟇 蟆企┯ 覲
op.gg覲企 螳 朱 蟆 (豕蠏 20 蟆)
蟯 ｌ企蟇 蟆讌
1 0 0 1 0
豈殊
0 0 1 1 0
豈殊
+
Team 100
Team 200
Sparse Matrix
<螳 襴>
+ 3 0 2 1 0 3 2 1 0 0
Team 100
螳 危伎
願下朱 + 1 譟朱 -1 * alpha ( < 1 )
豕譬 螳](https://image.slidesharecdn.com/random-180115015100/85/LOL-win-prediction-14-320.jpg)
![[NDC2016] TERA 覯 Modern C++ 蠍](https://cdn.slidesharecdn.com/ss_thumbnails/v09teramodernc20160425-160427044156-thumbnail.jpg?width=560&fit=bounds)
More Related Content
More from Dong Heon Cho (20)
LOL win prediction
- 1. 蠍 るゼ 襷ろ 1 豪Μ 豸 davinnovation@gmail.com
- 2. 一危 覿 覦蠍 一危磯ゼ 覲企 語, 語企 蠍朱 襾碁Μ 旧旭 覦覃伎 覦襦 襴糾 覿 襷豺螻 覈語 覦 豢 scoreboard 覲伎 螻 1 覃語 蟾 旧旭 襷 襾語 覈語 危 れ 覈語 蟲 一企 襦蠏碁覦 ル 豕 れ伎 豕 覦 豕 豐煙覲伎
- 3. 一危 覿 覦蠍 譬蟆 襷覃 Heuristic 讌れ朱 NOGADA Try-Error-Feedback
- 4. 覓朱 瑚 給 譬 瑚るゼ TIP 1. 襷 螻褐 == 一危 り 2. れ 螻褐 == 覈語 れ 3. 譬 一レ == 譬 Framework 覈 (*3 elice 讌ъ 蠍郁 覦 覃 襷れ 譴 .)
- 5. 襷 螻褐 == 一危 り { "matchDuration": 1716, // 豐 , 朱 蟆 る 讌螳襯 覩 "teams": [ // 2螳 Dictionary { "firstDragon": true, // (覈ろ)襯 襾殊 譯曙螳 "dragonKills": 2, // 覈覯 譯曙螳 "winner": false, // 豪Μ螳 "firstBaron": false, // 覦襦(覈ろ)襯 襾殊 譯曙螳 "baronKills": 0, // 覦襦 覈覯 譯曙螳 "firstBlood": false, // 豌 螳 "teamId": 100 // ID }, ... ], "participants": [ // 10螳 Dictionary { "championId": 412, // 豈殊 (貂襴) ID "summonerId": 21983, // 危 蟆企┯ ID "teamId": 100, // ID (teams teamId 襷れ) "stats": { // 企 糾襯 覩 "kills": 2, // 蟆企┯螳 るジ 蟆企┯襯 覈覯 譯曙螳 "deaths": 8, // 蟆企┯螳 覈覯 譯曙螳 "assists": 11, // るジ 蟆企┯襯 譯曙企 覈覯 譯殊螳 "goldEarned": 7314, // 豐 螻 (蟆 ) "totalDamageDealt": 27629, // 覈ろ + 蟆企┯蟆 螳 豐 磯語 "totalDamageDealtToChampions": 9507, // 蟆企┯蟆襷 螳 豐 磯語 "totalDamageTaken": 20419, // 覦 豐 狩 "minionsKilled": 37, // 譯曙 覩碁(覈ろ) "totalHeal": 1014, // 豐 豺 "totalTimeCrowdControlDealt": 241, // るジ 伎伎蟆 Crowd Control (CC) 襯 豐 螳 "wardsPlaced": 5 // Ward(朱ゼ 譯朱 危) 襷旧 レ逢 "items": [ // 危 覲 3401, 2049, 1031, 3270, 0, 2043, { "teams": [ // 2螳 給. { "teamId": 200 // ID }, ... ], "participants": [ // 10覈 蟆企┯ 覲 { "championId": 421, // 豈殊 ( 貂襴) ID "summonerId": 22082, // 蟆企┯ ID "teamId": 100 // ID }, ... ] }
- 6. れ 螻褐 == 覈語 れ And Hyperparameter Tuning
- 7. 譬 一レ == 譬 Framework
- 8. Experiment. Iter 0 { "teams": [ // 2螳 給. { "teamId": 200 // ID }, ... ], "participants": [ // 10覈 蟆企┯ 覲 { "championId": 421, // 豈殊 ( 貂襴) ID "summonerId": 22082, // 蟆企┯ ID "teamId": 100 // ID }, ... ] } Test 覦 覲企 豈殊 ID, 蟆企┯ ID, <螳 襴> 蟆 磯襯企ゼ 覦危讌 螻 譴 蟇 豈殊語願讌? 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 Team 200 Sparse Matrix
- 9. Experiment. Iter 0 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 Team 200 <- Data Input <- Model <- Output0 || 1 Logistic Regression, Decision Tree, Neural Network, Gradient Boost, SVC, KNN 願碓 朱誤磯 螻, 覈碁 る慨螻, 觚 襷 伎狩螻
- 10. Experiment. Iter 0 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 Team 200 <- Data Input <- Model <- Output0 || 1 螳 ろ伎 宴
- 11. Experiment. Iter 0 result 襦れ 蠏碁 貉碁， 蟆碁 豈殊碁 覲企 65 蟆讌? 蠏瑚 襷ろ1 碁;; 豈 レ碁 蟇郁 覓企Μ OP豈 企 螳 蟆讌 殊伎 螻襷 襴螳
- 12. Experiment. Iter 0 result 85??? 襦れ OP譟壱 蟆伎給も 覯曙伎語 襷譟 蠏碁り 朱.
- 13. Experiment. Iter 0 result
- 14. Experiment. Iter 1 { "teams": [ // 2螳 給. { "teamId": 200 // ID }, ... ], "participants": [ // 10覈 蟆企┯ 覲 { "championId": 421, // 豈殊 ( 貂襴) ID "summonerId": 22082, // 蟆企┯ ID "teamId": 100 // ID }, ... ] } 豈殊 覲, 覲企 譴朱 蟇 蟆企┯ 覲 op.gg覲企 螳 朱 蟆 (豕蠏 20 蟆) 蟯 ｌ企蟇 蟆讌 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 Team 200 Sparse Matrix <螳 襴> + 3 0 2 1 0 3 2 1 0 0 Team 100 螳 危伎 願下朱 + 1 譟朱 -1 * alpha ( < 1 ) 豕譬 螳
- 15. Experiment. Iter 1 3 0 2 1playerA B,C,D,E 3 -1 1 0playerB A,C,D,E 蠍 伎 譟壱襷 螻 . X 10 覿襦 : 覯 轟 豌 磯 覿襦 : 螻 碁，襷 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 +
- 16. Experiment. Iter 1 85 ( 讌 )
- 17. Experiment. Iter 2 500playerA 0playerB 企 豈殊 . X 10 蟇一 覃 豈蠍 豈 襴 || train れ 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 +
- 18. Experiment. Iter 2 85
- 19. Experiment. Iter 3 500playerA 0playerB 豈殊語 豪(train) . X 10 磯蠍 豈 OP 豈碁 1 0 0 1 0 豈殊 0 0 1 1 0 豈殊 + Team 100 +
- 20. Experiment. Iter 3 85
- 21. Experiment. Iter 4 Iter2 + Iter3 + Iter4
- 22. Experiment. Iter 4 86! Auto ML 覈語 Gradient Boost Model ( Not Ensemble Model )
- 23. Experiment. Iter 5 Iter2 + Iter3 + Iter4 渚覃 Conv1D ==
- 24. Experiment. Iter 5 87!!! れ 碁企 val_acc 豌 . Callback朱 豕螻 覈語
- 25. Abstract : ML 蠏碁 覃 覃 讌 襷 碁企 螳譬 瑚も 殊伎 豪 語 螻 蟾
- 26. 蠏碁 螳 覯 覓殊企瓦給 Q. 襷ろ 一伎 豪Μ 覓伎瑚? 譟壱襷 覃 覿覿
- 28. If 1譯殊殊 朱 襷 企 螳 螻 一危 譴 曙企慨螻 覿覃 譬給も
- 29. Q & A 給 伎 davinnovation@gmail.com