How To Upload Hearthpwn Decks To Hearthpwn

Purifying Hearthstone Data

Tabular array of Contents

1. Overview
2. Questions
3. Cleaning
4. Visualization
5. What'southward a Purify???
6. Return to Visualization
7. Conclusion
8. Photo and Data Credits

Overview

Hearthstone is a digital carte game released by Blizzard Entertainment in 2014. It's a free to play game featuring many prominent characters from World of Warcraft. Hearthstone quickly became extremely pop due to having like gameplay to Magic: The Gathering while also having a much lower barrier to entry. It has many facets that make it unique due to its electronic nature, such as:

• Creating random cards from a class and shuffling them into your deck
• Cards that replace your deck with a different ane mid-match
• A card that shuffles a copy of your opponents deck into your deck

It has many built in gamemodes, just the main mode, and the mode I'thousand looking at with regards to the data, is ranked play. The other modes are not as interesting to look at as many of them change weekly or involve playing solo.

The basics of a game Hearthstone are every bit follows:

• Each histrion has a xxx-card deck
• The deck type is chosen from 1 of 9 classes
• Each deck fundamentally contains a mix of two bill of fare types, minions and spells
• Minions are played on the board, and can be used to assail your opponent or their minions
• Spells are single use effects that can be used on a diversity of targets, such equally minions, your deck, your opponent, etc.

The dataset was pulled from a popular deck sharing website called hearthpwn, and contains over 340,000 decks uploaded from the game's release until February of 2017. The dataset likewise contains a .json file with every card in the game and their attributes.

Questions

• How did the popularity of Hearthstone modify from its initial release as they released more content?
• What classes were popular/unpopular for each content alter? Practise they correspond at all with what the highest rated decks were?

Cleaning

Its immediately obvious that there's a lot of bad data in here. The deck csv was a bit of a claiming. It started with 346,232 decks, with each carte du jour in the deck in it's own cavalcade. That means a whopping xxx columns just for the decklist. My start thought was to take all 30 card columns and combine their values into a tuple. That would allow for significantly less width on the dataframe and also easily allow for dropping duplicates afterwards. After that, I put each tuple'd decklist into a new column I titled "card_list". Now all that was left was to drop the original xxx columns that contained ane card each and move on. I was later on this I realized that the information also included decks from many of Hearthstone'southward special gamemodes, so I just wanted the decks that were made for ranked play. At present I've got a 202,375 row data frame of merely decks in the ranked play manner, it's done right? Wrong. Keeping duplicate deck lists didn't seem very important as there were many copies of popular deck lists that were uploaded with little to no ratings. I sorted the data by ratings and dropped all indistinguishable decks with lower ratings, bringing the concluding information size to 184,903 decks.

With the card json, there was also a lot of muddied data, but it was easy to decide what to continue. Since any single role player specific cards or even joke cards that you only see when you watch the credits of the game were as well in this file, it had a lot of cards that literally would never show up in a ranked deck. While I didn't need to trim this information, I wanted to make the frame as pocket-sized as possible to increase processing time. The most important thing I wanted to do was reduce the columns, as each card independent 31 columns of data. I was able to cut this down to 16 of import columns, which was later on dropping all columns with irrelevant or redundant data.

All of the code used to clean the information can exist found hither. The notebook used for plotting and data manipulation can be found here.

Visualization

What I really wanted to know almost this data (as a former player) is how balance updates to certain cards or new releases affect what is played or popular. As a starting point, this is the total corporeality of deck uploads per content update.

One could gather that the Explorers and Old Gods content releases were the nigh pop. While this might be interesting to wait as for someone who's played the game before, I want to also show how each of these compared to uploads throughout the overall timeline of the game.

This shows the total corporeality of uploads per content release compared to the overall monthly uploads. As the popularity of the game rises, the avg yearly uploads trends upward despite certain changes or expansions existence less popular than previous ones. This is most visible on the Explorers set, the nearly popular set as far as uploads are concerned, yet the largest upload months were the months around the Sometime Gods and Gagdetzan expansion releases.

Another thing I wanted to visualize was the most popular class and deck for each content update. While this might be the most interesting to people who play the game, information technology can give a absurd insight on how the meta adjusts when things are inverse.

Starting with the most popular form per patch:

And on to the highest rated decks.

It'due south pretty hard to tell at a glance how these line up, so let'south mark them based off of if the deck matches the pop course rather than only by class.

So there really doesn't seem to be any sort of necktie-in between what decks were highly rated versus what classes people like playing. But there information technology something interesting about the highest rated decks. Every deck on information technology is a deck with a high winrate, except for 1. This beautiful lilliputian outlier is Purify Priest. Purify Priest (at the time) boasted, if you were lucky, effectually a 10% winrate.

What's a Purify???

While this may non seem either bad or good, this menu was by miles the worst card that had been added to the game. It was so bad it became a claiming to try and brand a deck that could actually win centered effectually purify's effect. Fortunately (perhaps), that challenge took off and consquently created a whole cult of Hearthstone fans who called the Purify Priest the "Unicorn Deck". Something that's and so mystical and powerful that it couldn't possibly exist. The high rated version of it that is seen on the above graph is a remnant of this belief that it truly was the strongest deck, just no 1 knew how to employ information technology.

Return to Visualization

The highest rated decks after Purify Priest are actual skilful decks over again, but the post-obit behind this unicorn deck caused an interesting blip on an otherwise pristine chart. Funny enough, Purify only appears in 7% of Priest decks from its release until the next expansion came out three months after. Leading down an interesting idea of determining the nigh pop cards for each class in ranked decks. Let'due south look at the Priest cards start.

If you're angry that our lord and savior Purify isn't on this nautical chart, please click here for a version of it that says Purify for every card. If yous're not, let'south keep. Something of notation on this plot is the fact that some of the elevation fifty Priest cards are actually Neutral cards. I felt these were of import to keep in equally even though they may be Neutral, these are the specific Neutral cards near used in Priest decks. Each class has different Neutral cards that happen to synergize with their cards. Nosotros can get a proficient thought of which Neutral cards are shared amidst the virtually classes past looking at the top fifty cards overall.

That summit 15 right there is what I'm talking nigh. Those are the Neutral cards consistently detect their way into 2, 3, or even five different classes. Let'south peek at the top fifteen cards for each class.

Azure Drake shows up in v dissimilar classes, and in forty%-50% of decks for those classes. Sylvanas Windrunner, the overall 2d place shows upward in 3.

Conclusion

Information technology'due south pretty patently to see that whenever new content is released, the deck uploads fasten briefly and slowly decline until new content comes out once again. What makes the data nigh interesting to me is how easily it can be skewed. Just a single bad carte du jour made a bad deck "good" purely because people were making fun of it. Or a single expert card makes marginal popularity improvements go massive improvements. Hearthstone's popularity continues to rise even if the monthly submissions tedious briefly.

Photograph and Data Credits

• I did not create nor do I ain whatsoever images/information from the game of Hearthstone
• This data was collected from https://www.hearthpwn.com using their API by a Kaggle user named romainvincent. The data can be accessed by clicking hither.
• Hearthstone, Globe of Warcraft, and all related properties are owned by Blizzard Amusement
• The logo image was downloaded from the principal website for Hearthstone.
• The gameplay image was pulled from an commodity on https://www.windowscentral.com, the commodity tin exist found here.
• The image of the Purify card was pulled from the Hearthstone Wiki, the page about the card Purify can be found here.
• All graphs and charts were made past myself in matplotlob using the data from Kaggle

Source: https://githubhelp.com/adws2/Hearthstone-Data-Analysis

Posted by: bryantlosigiand.blogspot.com

Share this post