Timeline for Chess Analysis with Limited Information
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 17, 2020 at 9:04 | history | edited | CommunityBot |
Commonmark migration
|
|
| Mar 28, 2017 at 15:09 | comment | added | Nathan Merrill | Hey, I've added a ton more test cases if you are interested. | |
| Mar 24, 2017 at 7:58 | comment | added | justhalf |
Sorry I didn't reference my claim earlier. The 75% figure is based on my comment in the question, responding to xnor call for confirmation on the optimal score. An optimal system that outputs the true probability will get a score of 75% if p=0.5 as can be seen in the plot I gave in the comment. If p!=0.5 (which is likely here) then the optimal score will be higher than 75%. And yes, the comparison is not fair with user1502040 since more data is used there.
|
|
| Mar 24, 2017 at 7:20 | comment | added | Fatalize | @justhalf This is exactly why (1) I did not use 100% of the set to train (2) have commented on OP's question to tell them that they should use a secret set of data to test all answers. I don't know how you can pull that 75% percent figure out of thin air. user1502040 uses 90% of all data for training instead of 80% as I do. | |
| Mar 24, 2017 at 6:45 | comment | added | justhalf | The 94.5% is the result on the training data? I don't think that number is meaningful as neural network also works as some kind of hard-coding mechanism if the parameter space is large enough. Note that if the probability that white wins is close to 0.5 (which is quite true if we are not given any information), the optimal expected score should be only slightly larger than 75%, which your result on test data seems to match quite well, but not the result on training data. The result by @user1502040 seems more reliable (on validation set) and realistic. | |
| Mar 23, 2017 at 19:25 | comment | added | user1502040 | How does weka know you are trying to predict the winner? | |
| Mar 22, 2017 at 13:08 | comment | added | Fatalize | @NathanMerrill Edited with a paragraph on the feature vector construction. | |
| Mar 22, 2017 at 13:07 | history | edited | Fatalize | CC BY-SA 3.0 |
added 577 characters in body
|
| Mar 22, 2017 at 13:03 | comment | added | Nathan Merrill | How are you passing the test case as input to the neural network? Are you just passing in the raw string? | |
| Mar 22, 2017 at 12:59 | comment | added | Fatalize | @NathanMerrill I'm not sure I understand your question | |
| Mar 22, 2017 at 12:59 | comment | added | Nathan Merrill | How do you encode the input? | |
| Mar 22, 2017 at 12:56 | history | answered | Fatalize | CC BY-SA 3.0 |