Introducing cricodds: Match Odds and Analysis

[viriya]

Another reason why I think it should be based on RRR deviation:

Yea probably better to find similar based on run rate (req rate for 2nd innings) overs and wickets instead of runs, overs and wickets to lower the chances of the edge cases you pointed out earlier. I'll backtest this to see if the predictions are better.

 

[weldone]

Yea probably better to find similar based on run rate (req rate for 2nd innings) overs and wickets instead of runs, overs and wickets to lower the chances of the edge cases you pointed out earlier. I'll backtest this to see if the predictions are better.

Wickets are fine. Just to make it clearer - you are currently doing it on the basis of (runs and wickets for overs in range). My suggestion is to do it based on (RRR and wickets for overs in range).

 

[viriya]

Wickets are fine. Just to make it clearer - you are currently doing it on the basis of (runs and wickets for overs in range). My suggestion is to do it based on (RRR and wickets for overs in range).

Currently it's runs in range, overs in range, wkts in range. It will be rr/reqr in range, overs in range, wkts in range.

 

[weldone]

Currently it's runs in range, overs in range, wkts in range. It will be rr/reqr in range, overs in range, wkts in range.

Yep that's what I meant.

 

[viriya]

Also, why are there no 2nd innings records with exactly 50 overs remaining? (it starts from 49 in your site - should also include lunchtime data if possible)

Actually this isn't an issue. Because if it's 0 overs bowled in the second innings you can just input 1st innings data with 50 overs bowled to get odds for the 1st innings batting team instead. The odds won't work as well after the 1st over in the second innings, but I don't think it's big enough of an issue to focus on atm.

 

[weldone]

Actually this isn't an issue. Because if it's 0 overs bowled in the second innings you can just input 1st innings data with 50 overs bowled to get odds for the 1st innings batting team instead.

No because number of wickets after 50 overs don't matter. The only way is to check the odds for all possible wickets combinations (0 to 10) and take an average.

 

[viriya]

No because number of wickets after 50 overs don't matter. The only way is to check the odds for all possible wickets combinations (0 to 10) and take an average.

If you think in those terms something of that sort would need to be done for the 49th over as well since the value of wickets goes down with overs.

I'd have to look at the data a bit to figure out some sort of function for the wicket range to consider vs overs - would add some complexity though.

 

[weldone]

would add some complexity though.

good analytics is often complex

 

[viriya]

good analytics is often complex

Yes but my goal was to make a tool that was usable and easy to understand.

Also there is an argument that if a team is 250/5 after 50 overs vs 250/10 even though it's the same score it's not the same scenario because more wicketo suggest that wicket taking is easier. So might sense to keep to a range of similar wickets still.

 

[weldone]

Yes but my goal was to make a tool that was usable and easy to understand.

Also there is an argument that if a team is 250/5 after 50 overs vs 250/10 even though it's the same score it's not the same scenario because more wicketo suggest that wicket taking is easier. So might sense to keep to a range of similar wickets still.

Yes, and there are other complications - we can discuss that later.

The RRR change seems to be the most needed one atm.

 

[viriya]

Yes, and there are other complications - we can discuss that later.

The RRR change seems to be the most needed one atm.

I did some tests using RR/RRR ranges instead of runs over the weekend without great results. I think the problem is when the RR/RRR gets too high or low you range needs to adjust accordingly.

 

[weldone]

I did some tests using RR/RRR ranges instead of runs over the weekend without great results. I think the problem is when the RR/RRR gets too high or low you range needs to adjust accordingly.

Does +-10% RRR range work? In fact, personally I'd also suggest combining the RRR and wickets ranges. Example below:

Query: 8 RRR runs with 6 wickets left?
Results:
1. Cases with 8-8.8 RRR with 6-7 wickets left
2. Cases with 7.2-8 RRR with 5-6 wickets left

The overs range can stay as usual

 

[viriya]

Does +-10% RRR range work? In fact, personally I'd also suggest combining the RRR and wickets ranges. Example below:

Query: 8 RRR runs with 6 wickets left?
Results:
1. Cases with 8-8.8 RRR with 6-7 wickets left
2. Cases with 7.2-8 RRR with 5-6 wickets left

The overs range can stay as usual

My test was with +-10% which didn't give conclusively better results. I tested a combination of RRR/RR and RunsReq/Runs depending on how high the RRR/RR was already and it had mixed results.

I've been trying to find a proper measure that combines RRR and wickets already.. Just plain RRR/Wkts is too trivial since the value of wickets diminishes the higher the number.

 

[viriya]

Actually I revisited the unique match scenario change and turns out it doesn't actually improve the odds even though it seems logical. More scenarios even from the same match turns out to give better odds.

 

[weldone]

Don't be over-reliant on backtesting. Follow the sound logic more than backtesting results, unless the results are way way worse.

 

[weldone]

Also, I can't believe you're not getting better results after using RRR. I can clearly see from the data that from the same query the lower RRR generally have higher win ratios.

Example:
Just look for 70 needed off last 6 with 4 wickets in hand from 4-Oct-2012...

ALL wins come where RRR < 10
MANY losses are from the RRR > 10 region

So, in this case the 45.45% win odd is wayyyy overstated, because there is not a single win where the RRR is equal to or greater than the actual (11.66). Also, no wins till 10% below the RRR.

This is just one example. All such scenarios give similar results.

There may be something wrong in your backtesting methodology.

 

[viriya]

Also, I can't believe you're not getting better results after using RRR. I can clearly see from the data that from the same query the lower RRR generally have higher win ratios.

Example:
Just look for 70 needed off last 6 with 4 wickets in hand from 4-Oct-2012...

ALL wins come where RRR < 10
MANY losses are from the RRR > 10 region

So, in this case the 45.45% win odd is wayyyy overstated, because there is not a single win where the RRR is equal to or greater than the actual (11.66). Also, no wins till 10% below the RRR.

This is just one example. All such scenarios give similar results.

There may be something wrong in your backtesting methodology.

Thing is, using multiple scenarios from a match is not necessarily illogical - it's just unconventional.

RRR performing worse is probably partly because i'very only tried +-10% so far and not RRR-wkts.

 

[weldone]

Thing is, using multiple scenarios from a match is not necessarily illogical - it's just unconventional.

RRR performing worse is probably partly because i'very only tried +-10% so far and not RRR-wkts.

There are some serious problems man!

I am querying this scenario: T20, runs required 115, overs remaining 8, wickets 0.
So, the RRR is 14.375. Ok?

Now, the results: 165 similar count, 49.09% win odds!. I was a bit surprised to see this high win odds in such a scenario. So, I look at the results:
Among 165 'similar results' only 1 has RRR greater than or equal to 14.375 and that is a loss.

Do you really believe teams can chase down 115 in 8 overs successfully half the times????

This surely needs fixing.

 

[viriya]

There are some serious problems man!

I am querying this scenario: T20, runs required 115, overs remaining 8, wickets 0.
So, the RRR is 14.375. Ok?

Now, the results: 165 similar count, 49.09% win odds!. I was a bit surprised to see this high win odds in such a scenario. So, I look at the results:
Among 165 'similar results' only 1 has RRR greater than or equal to 14.375 and that is a loss.

Do you really believe teams can chase down 115 in 8 overs successfully half the times????

This surely needs fixing.

I'm talking ODIs right now - haven't tested T20s extensively yet.

 

[weldone]

I'm talking ODIs right now - haven't tested T20s extensively yet.

same issue for ODIs for cases with high RRR and low number of overs remaining...the win odds are overstated and almost no wins at or higher than mentioned RRR