Do you like pointless data? Great! Let's look at karagiri rates in Houou games! If you don't know what "karagiri" means, it's when you draw a tile that you already have in your hand, then cut that tile from your hand. For example, you draw a 7p, then cut the 7p from your hand instead of the one you drew. Cutting the one you drew is called "tsumogiri." This can have some mind-games attached to it, theoretically, if people are watching your discards. Anyway, let's begin by simply looking at how often people karagiri at the different shanten numbers. For this, any time a player who isn't i ..read more

On Amae Koromo, the stats aggregator for Mahjong Soul, it lists the average starting shanten in Jade room as being around 3.482. People thought this sounded low, so let's see the averages in Houou. Let's start by simply looking at the average shanten of all the starting hands in Houou Hanchan games. Tenhou's wall code is public and we know it's fair. This is a high sample size, with 1.5 million games. The numbers won't deviate much. For example, here's the same chart for East-only games, which has a sample size of half a million. It's only different by roughly 0.0003. These numbers are a fu ..read more

The developers of Naga have released a paid service where you can have Naga review your games. Let's go over it! You can find the information page (in Japanese) here. The purchase page is here. There are three tiers of pricing. Each replay review costs 50NP. Gold Tier - 550 Yen per month, 500NP (10 reviews) per month Platinum Tier - 1100 Yen per month, 1500NP (30 reviews) per month Diamond Tier - 2200 Yen per month, 5000NP (100 reviews) per month The NP doesn't carry over between months. If you have excess near the end of the month, consider asking in the Riichi Discord whether anyone has a ..read more

In this analysis, we'll look at several aspects of the game and see how they affect whether a player chooses to riichi or dama a tenpai hand. In the previous post, we saw that, overall, riichi and dama have similar win rates. One factor to this could be that people choose to dama in more dangerous situations, making other players more likely to win. One easy measure of how close someone is to tenpai is how many times they've called. Let's see how the rates of riichi and dama change based on the number of calls in the game. Aside from 10 and 11 calls, which both have only one instance each, t ..read more

I'm finally analyzing dama hands with the help of the drastically faster shanten calculation algorithm. Let's take a look at these mysterious creatures. My first instinct was to replicate my old Wait Win Rates by Row analysis with dama hands. However, dama hands are very complicated. Let's think about what a player with a dama hand could do. Stick with their wait to the end of the game. Change their wait, but remain tenpai. Fold at a later point. Declare riichi later. Call a tile, either to widen the wait with kuinobashi or to go down a folding path. Pass a win because they need to target a ..read more

In this post, we'll look at the chance someone is tenpai based on the number of tedashis they've done, as well as what that tedashi is. Before we get into the analysis, I want to mention something. This analysis involves calculating shanten a lot. A LOT. The analysis has to calculate the shanten of a hand over six hundred million times to gather this data. This would be highly impractical with the usual recursive shanten algorithm. I ported over an algorithm written by spinesheath. You can find it in C# here, and my python conversion here. This algorithm is much much faster. It reduced the ti ..read more

In this post, I'll go over the scripts I use to analyze replays, for anyone who also wants to try to do some analysis. Naturally, you'll need to know some amount of programming. We'll also look at a silly analysis I did which looks at the average composition of discards based on the hand's yaku. First, you'll need the actual replays. You can use the phoenix-logs repo to do this. However, I had trouble getting it to work, so ApplySci sent me the database files themselves. I've put five years of replays in this Google Drive folder you can download. It should be a good enough sample size for most ..read more

Today we'll be talking not about toilets, but about honitsu or chinitsu hands. It's often said that a flush hand becomes particularly dangerous once they cut the first tile of their suit (known as overflow), so let's dive into the Houou replays and see what the data says. My dataset has expanded to include games up to the end of 2020, increasing the sample size to 1,539,555 replays. In these, there's a bit over four million instances of an overflow occurring that the analysis captures. We'll only be looking at open hands, since it's easier to know whether they are going for a flush. We'll als ..read more

Do the stats of a Houou player have any correlation with each other? Does someone who calls more win more? Do they deal in more? Have lower value? Let's look at a bunch of these. The data used here is the same as from my The Average Houou Player post, filtered to only include players with more than 300 games to avoid wild data from low sample sizes. This leaves 3590 players. The first thing we'll look at is how call rate correlates to win rate. Does having a higher call rate result in a higher win rate, and if so, how much? Let's quickly break down this chart. Each dot is a player, where the ..read more

Someone sent me a blog post with information on the average shanten after a player makes a call, then asked if the tonpu numbers would be different. So, I recreated the code and went into more detail. What's the average shanten after someone makes their Xth call on Y turn? What about their chance of being tenpai? Does the type of call affect it? Let's see. You can find the original blog post here. I'm not entirely sure on their methodology, but it does say the data was from the 2016 replays. My data set is around 8 times as large, so there will be some number differences. As for my methodolog ..read more