Dev Insights: Treasure Pools & Smart Loot Buckets in Duels

Hey there! My name is Ates and I’m an Associate Game Designer on the Hearthstone team. The team is thrilled to see everyone jumping into the new Duels mode. So, I wanted take opportunity to fill you all in on how getting Treasures in Duels works!

There are two different types of Treasures that we offer:

• Passives: Treasures that offer a benefit for you all game. A player can obtain a maximum of 2 of these Treasures during their run.
• Actives: Treasures that are additional cards that get shuffled into your deck before the game starts. A player can obtain a maximum of 5 of these Treasures during their run.

Like Dungeon Runs of the past, we split our Treasures into these two categories. To control power levels and spikes throughout a run, we further divided both Passive and Active Treasures into different pools. This allows us to better control the impact each Treasure card has.

In Pool 1, you are likely to see more simple Treasures. The goal with this pool is to offer a choice that helps the power of a deck and potentially gives an option for the player to build around, without letting power levels get too out of hand.

In Pool 2, we turn up the heat a bit. The cards you see here should be more powerful than any hard synergy in your deck, with a few exceptions. Within Pool 2, there are some Treasures in a category called Ultra Rare. Ultra Rare Treasures are the most powerful Treasures of all! Whenever you are offered a Pool 2 Treasure there is a 1% chance you will see one from the Ultra Rare pool.

We use these pools as a mechanism to help us buff or nerf a Treasure's relevance by moving it between different pools. As an example, you may remember Party Portal lost some health in a recent balance patch. At the time of that adjustment, Party Portal was in Treasure Pool 2. After looking at the card, we decided that it was a better fit if we shaved off a bit of health and instead offered it as a Pool 1 choice.

Here is when you will see Treasures for each pool.

Passives:

• Pool 1: After game 1.
• Pool 2: After game 2.

Actives:

• Pool 1: After games 3, 9, and 13.
• Pool 2: After games 7 and 11.

Finally, here is a list of all available active and passive Treasures we offer, along with what pools they are available in!

Group Learning Loot Buckets

Greetings! I’m Tian, a Lead Data Scientist on the Hearthstone team. I’m going to talk about some science behind the loot buckets in Duels! When you’re offered a loot bucket between games in your Duels run, you might be curious about the Group Learning bucket, which is always in the middle. Let me explain what that is, how it’s generated, and how it’s recommended to you!

What is a Smart Bucket?

We also call Group Learning buckets “smart buckets”, or auto-generated buckets. Compared to the other two buckets, hand-crafted by our talented design team, smart buckets are generated automatically from big data. Big data utilizes gameplay data we’ve been gathering since the early days of Hearthstone across the Standard and Wild ranked modes, and Arena mode, to generate those smart buckets and to recommend them to you!

As a side note, we have similar smart technology being used in the auto deck builder in your collection manager as well – please check this previous blog if you are interested.

How the smart buckets are made?

We first define “synergy score” between card pairs, which is learned from our enormous gameplay data. For each feasible card pair, synergy score looks into:

1. How were cards played in each game?

In simple words, if we see two cards were played by players repeatedly, it’s a good indicator that there is some synergy between them.

2. Which stage of the game and what turns were these cards played?

Cards played in close turns and turns far apart usually indicate different things. The previous pair usually shows signs of combo pieces, and the latter shows cards used as earlier tools and late-game powers -- for example, Arcane Missiles and Flamewalker vs. Doomsayer and Alexstrasza. We have knobs in the score formula that controls how we evaluate such turn information and as a result, we have many different versions of scores based on how these knobs are being turned!

3. What is the end-game impact (i.e. win/loss) after the card pair was played?

Win-rate information is embedded in the card pair synergy score. In general, the higher the win-rate, the bigger the score.

After we have pair-wise synergy score, we then aggregate them to get the synergy score for buckets, with some restrictions applied – for example, cards in the bucket need to be different. Since each bucket contains 3 cards, {A,B,C}, the aggregation consists of 3 different card pairs: {AB, AC, BC}. Mathematically speaking, we choose a “reversed soft-minimum” function as our aggregation method. The benefit of this over simple addition or some other aggregation methods is it helps to avoid the scenario that {AB} and {AC} has a large score value, but {BC} has a very small score value.

How buckets are recommended to you?

A pool with millions of feasible smart buckets are updated daily (pre-generated) and stored in a server’s memory cache for efficiency. The decision of which bucket is shown at each loot bucket selection moment is a dynamic live process, as it is determined by your hero class and your current deck content.

During this live recommendation phase, we calculate a synergy score between any bucket of interest and your current deck. Here are the steps in details:

1. We compute the initial deck-bucket synergy score with “reversed soft minimum” aggregation.
2. We apply some math adjustments, such as normalization, applying additional functions, etc., to the score.
3. We adjust the score by checking your deck’s current Mana curve – for example, if you already have a lot of low Mana cost cards in the deck, we will prioritize buckets with cards of higher Mana cost.
4. If your deck already has one or more of the same card(s) in the bucket, we will also apply card multi-appearance penalties to the score – the more duplicates, the bigger the penalty is. For example, if you already have 5 Fireballs in your deck, the chance of you seeing another Fireball in the smart bucket is very low.

Once the synergy scores are calculated for each bucket, we sample one bucket by the synergy score value – that is, a bucket with higher score is more likely to be picked than a bucket with lower score. This is a process with some “controlled randomness” involved: we do not always give you the highest score bucket! By doing so, we hope there will be more diversity and fun in your deck building process through the run.

I hope this blog helps you to better understand our data-driven approach in Duels. The smart bucket system has been set up to be as automatic as possible, and it evolves as Hearthstone evolves. We will continue to iterate and improve on smart buckets where we can. Currently, hero powers and treasures are not included in the “learning” process, thus we are not able to tell the synergies between them and all other cards. However, we have plans to have them included in the future.

We hope you enjoy Duels! We are constantly listening to your feedback -- feel free to share any thoughts and comments about loot buckets and Treasure Pools with us. We’ll see you all in the Tavern!