Search

  1. 1
  2. ...
  3. 2
  4. 3
  5. ...

Forums


  1. Read the TrueSkill2 paper to see how we evaluate metrics like K/D and DPM and more examples. The main thing is, we don't include K/D or DPM in the model right now. If, for example, high K/D was important, then we would see TrueSkill2 underestimating...

  2. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  3. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  4. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  5. You are both misunderstanding win%. Raw win% is meaningless outside the context of who you played against. In fact, the raw win% and KDA values shown on the leaderboards on this site are also meaningless because they don't show who the opponent were...

  6. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  7. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  8. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  9. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  10. Having negative games doesn't change winnability in the data. Those players are still winning as predicted, whereas when we tested DPM it only made accuracy worse. So why would we change it? It's more accurate this way, and the players we look at...

  1. 1
  2. ...
  3. 2
  4. 3
  5. ...