Verified Commit 2a1dfbb3 authored by Markus Shepherd's avatar Markus Shepherd 🙊
Browse files

How good were the predictions?

parent 4d5bce48
......@@ -120,3 +120,29 @@ Not to brag, but I called this one [months ago](
* {{% game 251247 %}}Barrage{{% /game %}}: Definitely on the *very* heavy side even for a {{% kdj %}}Kennerspiel{{% /kdj %}}, yet [Udo Bartsch's high praises]( left no doubt about a recommendation.
* {{% game 291457 %}}Gloomhaven: Jaws of the Lion{{% /game %}}: Similar to the previous entry, I had a good feeling about a recommendation, but didn't think it'd be actually considered for the award.
* {{% game 291859 %}}Riftforce{{% /game %}}: I completely missed this one, as did the algorithm, placing it at position 81. At least it classified it correctly as a {{% kdj %}}Kennerspiel{{% /kdj %}} – with 51% confidence. 😰 The only pure two-player-game on either list, this proves how hard it is for lower player counts, even during a pandemic induced shortage of playmates.
# How good were the predictions?
This really is a drama in three acts: the decision of {{% kdj %}}Kennerspiel{{% /kdj %}} or not, the algorithm picking the longlist, and finally my guts picking the shortlist.
## {{% kdj %}}Kennerspiel{{% /kdj %}} score
When I introduced the [{{% kdj %}}Kennerspiel{{% /kdj %}} score]({{<ref "posts/kennerspiel/">}}) a couple of months ago, I didn't really have a good validation set to evaluate my model against. Now I do!
Out of the fifteen games on the longlist, the algorithm classified thirteen correctly. One was predicted wrong on either side of the line: The model saw {{% game 223040 %}}Fantasy Realms{{% /game %}} as a {{% sdj %}}Spiel{{% /sdj %}} and {{% game 317311 %}}Switch & Signal{{% /game %}} as a {{% kdj %}}Kennerspiel{{% /kdj %}}, but the jury saw it reversed. 87% accuracy is actually pretty decent for such a simple model trained on a small dataset, but it's still a nagging feeling just how far off the score for {{% game 223040 %}}Fantasy Realms{{% /game %}} was.
Anyways, we live and we learn, and the model has already been updated with the jury's latest decisions, so hopefully next year it'll work even better!
## Longlist
## Shortlist
All that's left to do is waiting. The jury will announce the winners on July 19, we'll be back shortly before that with the favourites for the two awards. 🤓
\ No newline at end of file
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment