Tactics Trainer Problem Difficulty

I am at around 1250 and the same thing has happened to me.

Yes, it is definitely a mess again. Why can't they test this on some test system and implement it once it is working again??

same thing

Is the algorithm for rating adjustment public? If not, why not?

There are a lot of smart people on chess.com that may see the flaw in the algorithm and can help to fix it.

I've reset my tactics rating to 1200 and won't be using the TT again until various problems are fixed. There are at present mch better sites for tactics.

Also, if this notorious member , who has an online rating of ca. 1600 (and a host of basically lost games being dragged out to keep his rating artificially high) can raise his tactics rating from around 1600 to 3100+ within less than two hours' work after getting his Diamond membership, there's something fundamentally wrong with the system.

I hope chess.com gets the problems cleared up soon. Until then, bye-bye TT for me.

Hey, guys... The logic to pick tactics for a given user was updated right around Christmas, so if there are still issues (observable within the last week or so), can we get some specifics?

The current logic isn't meant to be a secret. :) Here's how it's supposed to work:

X = your rating + 10

Tactics Trainer tries its best to serve you problems (you haven't already seen) within 50 rating points of X.

It is impossiblev to progress from say 1800, with time penalties and low returns on solved problems.

It is just a mess

Yeah, I've also noticed that points for getting tactics right are suddenly (well more than a month) very low. I used to be pretty consistent 1900 and on good days I'd get to 2000 (with a max of 2100). Now I'm stuck on the 1800 level, and on good days I almost reach 1900 ! Just because the points have gotten much lower for solving them. (+7 is about the max now)

Be it known (in the interest of full transparency), that the changes made in December (see post #196) are still in effect: The server is preferring tactics in a range from 60 points higher to 40 points lower than your current rating.

To date, although we have not changed the scoring logic, this change to tactics selection alone would generally reduce the size of ratings swing, whether you get a given problem correct or incorrect. (In other words, you're not going to see many ratings changes of +/- 22 if the problem's rating is proximate to your own.)

So that's where we are today...

However, with the advent of V3 later this year, we will be making another change to tactics logic (scoring this time) that we think most people will like (based on a recent survey). Here's what that will entail:

- Percentile score remains at 100% until the player uses up half of the problem's average time-to-solve. (This is an extension.)

- From that point on, the score drops steadily, reaching 50% at twice the average time-to-solve. At that point (a theoretical "draw" vs the tactic), the score stops dropping based on time. (It currently does this only after reaching 20%.)

- Any time after that, if the player solves the tactic correctly, the server will assess whether the player should get any points for a 50% score (i.e. if the tactic is rated higher than the player). If that's the case, then the player gets those rating points. In all other cases, the player's rating does not change. A player's rating will never drop as a result of correctly solving a tactic.

Thank you. Losing points for a correctly solved problem does not promote good chess. And it's annoying as heck.

That sounds like a really good update. Then the only thing I can complain about is that I think you should get 0% score when you don't get all moves correct.

Agreed. Getting 2/3 for dropping your queen on the final move is nonsensical.

. A player's rating will never drop as a result of correctly solving a tactic.

Can't wait! thanks.

Too late

Dear Admins

I took a timeout from TT for about 2 months now because of the bug I am going to write about again. I come back and see that if you set TT to a rating range of say 900 to 1300 and "unrated", TT will only give you problems at the high end of the range. Which is totally not what you want, in this case.

This used to work correctly in the past. Why is it not working anymore?? Has this feature essentially been scrapped? This I find actually much more annoying than dropping points for being slow solving a problem.

It was so nice, in the past you could set it to unrated at a range of say 800 to 2500 which would effectively give you just random puzzles. But this in my opinion was the best training because this is how it happens in real games. You never know the rating of a given position. Anyways, please kindly make that working again. It was there in the past.

Thank you for some answers.

I like these changes alot. it will discourage blitzing out quick guesses, and greatly encourage accuracy.

Is there a new bug in tactics trainer and is this related to the change of the problem rating distribution?

I have a rating of 1600 but the problems become easier and easier...

Three days ago the average problem rating for me was 1550

Two days ago the average problem rating for me was 1400

Yesterday the average problem rating for me was below 1100 (lowest 777, highest 1182).

I have a total of 1800 attempts and this should be far from a point where there are no problems left for me in that rating region. However, the distribution of the problem ratings changed from almost discrete with Gaussian envelope and average of about 1400 to a continuous distribution with an average of slightly below 1000 and a maximum at 600...

What happened? 

@Josimar73 - Does it seem like things have returned somewhat to normal recently? Well... if I look at your recent history, it seems like your rating has averaged around 1633 throughout the last 25 puzzles. During that time, the average rating of the puzzles you faced was ~1515. I would certainly agree that's lower than would be expected. But the *really* low puzzles have not shown up for a day or two. Anyway, I'll look into it.

@benkku52 - I'd love to see the changes delivered prior to the website overhaul for v3, but so far, it doesn't look like that's going to happen. At this point, I don't have a good ETA to give you. When I know something more specific, I'll post it here.

@DeepGreene

have you got any information on the issue/feature to set the TT to "completely random" by setting a very wide range and do "unrated" training? This used to work in the old days.

Even setting a narrow range say 1200 to 1400 and "unrated" does not work properly. TT keeps bringing only puzzles at the high end of the range. In the old days he would select puzzles from the whole range.

Or maybe setting a range manually and doing "unrated" will be abandoned entirely in the new version?

Thank you if have any information on this.

Hi DeepGreene,

Thanks for your feedback. Indeed the problems I get are in a "normal" rating range again. 

Best regards and thanks again,

Josimar