52 Bots

A small mechanical menagerie Christa Hartsock and Jim Fingal. Christa: web, twitter, botlist Jim: web, twitter, botlist.

[52bots]

#14 Rover Lapse

What:

Bot that keeps up with @earthroverbot (by @muffinista) and generates animated GIF summaries of where it has been.

How:

Runs every 4 hours. Pulls the images from eathroverbot, filters out duplicate images, condenses the images a bit, then uses the images2gif library to generate an animated gif.

I’m not super happy with the quality of images, but Twitter has an image size limit of ~3 megs and if I left the images at 600x600px I would only get about 10 in the animation -- crunching down to 350 lets me put 25 in there, but it looks kind of bad since the twitter gif viewer stretches the images to fit a standard size when it plays them.

I recently started a job where I work in Python all day and do a lot of unit testing, so this is the first bot I’ve made that is moderately well-tested. (I got slack towards the end when the finish line was in sight.)

Why:

I really like the concept earthroverbot but I don’t actually like following it in practice because it spams my twitter when Christa gets into moving it around, so I wanted something that condensed the experience.

Image editing has always felt a little magical to me (part of why I did tiny image bot), but after seeing the relatively straightforward source for @starnearyou by ddbeck​ I wanted to try it. My implementation differs in that I don’t use subprocess / I use a native python library to make the GIFs -- probably at the expense of quality, but fun to implement.

Source:

https://github.com/jimmytheleaf/roverlapsebot

[52bots]

Oh also, from this point forward, we’re basically only making @earthroverbot-related bots. The blog is now 52earthroverbotbots. sorry not sorry

[52bots]

Change of Pace

So, this may come as a shock, but 52 bots is scaling back the pace, which, you know, fair, considering the project was born of a conversation that was basically “I’ve never made a Twitter bot before!” *makes first bot* “Omg this rules! We should totally each make one a week the rest of this year!”

Anyway! Jim and I are still going to continue making bots and will post write-ups here, but the aim is now is quality > quantity. 

I personally was feeling a little burnt out leaving a trail of half-baked ideas and buggy bots in my wake, while having to punt more interesting but larger scoped ideas down the line due to totally arbitrary, self-imposed time constraints. 

Going forward, we’re both aiming to give these botbabies the time, tests (gasp!), and ideas they deserve, as they naturally occur.

!!BOTS!! \( ﾟヮﾟ)/

[52bots]

#13 You Are Here

What:

Bot that, given the query “where am I?” in a location-tagged tweet, returns a picture of the ground beneath your feet. Never look up from your phone again! Made for the Stupid Shit No One Needs and Terrible Ideas Hackathon, SF 2015.

How:

Listens for appropriate mentions on the Twitter Streaming API, then sends geographic coordinates or name of location (whichever is available) to the Google Street View API. Pic of the ground is retrieved by sending a low pitch as a parameter.

Why:

Misanthropy, mainly. Also, lols.

Source:

https://github.com/hartsick/you_are_here

Pitch Presentation

[52bots]

Due to bot forepeople’s relocation from LA to SF, bot production is on temporary hiatus! Factory output will resume in mid-April.

[52bots]

#12 LA Photo Bot

What:

Bot that pulls images from the Los Angeles Public Library's online photo collection and tweets them ~4-5x / day.

How:

Using seeded terms, I scraped results for a few searches on http://photos.lapl.org using Selenium. Some are entertaining and diverse ("dog" yields well, dogs, but also Pinks hot dog stand and a postal worker strike, among others), while others are more pointed ("protest", "oil", "abortion"). 

The bot then pulls a random entry from the database of scraped data, uploads the image file to Twitter's servers so that an image card is created, and composes a tweet with the image's title, date, photographer and collection (when available), and a link to the original record. 

Because of copyright concerns (the LAPL allows free personal use of images, but charges a fee for commercial use), the account is protected so that it stays small-scale. If you're interested in taking a look, though, please send it a follow! 

Why:

Because the LAPL's photo collection is an incredible resource that I've gotten a lot of joy and knowledge out of since discovering, but I only periodically visit the site—partly due to how difficult it is to browse. Really wanted a new way to explore items in the archive (and share them with others, but that's put on hold for the time being).

Source:

https://github.com/hartsick/image-archive-bot

[52bots]

#11: EBook of Black Earth

What:

An ebooks-style bot that makes extremely metal declarations.

How:

Most of the work for this was spent assembling the corpus and doing data exploration and trying to get the tweets to come out sounding interesting -- the algorithmic work with Markov chains had basically been done with Hwaetbot; I ported this and a number of other utilities into a botutils package which is pip-installable.

I used Scrapy to crawl a popular metal lyrics website and got lyrics for about 40k albums. Then, I find songs that contain at least one term in a whitelist of terms from Norse mythology, store that in an accessible corpus in Mongo. (I use Mongo since pragmatically I get more free storage on Heroku with Mongo than Redis.)

When generating the tweets, I feed the corpus into either a 2- or 3-gram (randomly chosen) Markov chain, then generate sentences until I get back an appropriate sentence that contains a word from the whitelist. It doesn't take long at all to train the model, so I just have this in a script that is run once an hour by the heroku scheduler add-on.

The MD5 digests of all original sentences and all new tweets that are generated are stored in a Redis set to avoid duplication. 

Why:

Inspired by a tweet exchange with deathmtn and inky. 

Source:

https://github.com/jimmytheleaf/aesirbot

(There are so many good names that I refer to this same bot with many different names.)

[52bots]

#10: rick r

What:

Bot that Rick rolls popular hashtags.

How:

Looks for trending hashtags in a random location. Ignores any promoted hashtags, or ones including "RIP"... and probably more things in the future. From this hashtag, it pulls a number of tweets (20-100) to use as its base. From this corpus, it then generates an eBooks-ish tweet of appropriate length, and appends both the original hashtag and a shortened URL of Rick Astley's Never Gonna Give You Up, generated using the Google URL Shortener API.

In the future, it may randomly change its name to shake people off and/or give Rick rolls as responses to its followers (a la Appropriate Tributes or the #wowwwnetwork ).

Why:

Because it's 2015 and Rick rolling still rules.

Also, less importantly, seemed like a good introduction to working with Markov Chains.

Source:

https://github.com/hartsick/rickrollbot

[52bots]

#9: Return Tea Table

What:

A bot that generates random table-flippers, and flips things, such as the OED word of the day and follower names.

How:

For flipping, has a table mapping characters to upside-down characters that it makes use of for flipping things. For flipper, has a number of possible features that get randomly selected and combined. In order to find things to flip, queries the OED homepage once a day and tries to parse the Word of the Day, skipping words it has already flipped.

Why:

Curious about the upside-down unicode characters, and wanted experience working with them. Like the cute little flipper guys.

Source:

https://github.com/jimmytheleaf/tableflipbot

[52bots]

#8: Best Intentions

What:

Bot that collects and shares new years resolutions, aspirations, hopes, frustrations.

How:

Listens for any replies to the bot in the User stream. Then searches the reply's text for commands: if the text includes 'in 2015' or 'I fucking did it', retweets and responds directly to the user with an affirmation. If the text includes 'help', it responds with a tweet to the requester with a suggestion for use.

Why:

2014 kind of sucked. Wanted to give a platform where a small, possibly ephemeral community could form around (potentially positive) plans, frustrations, and ideas. Also wanted an excuse to improve and refactor the Streaming API-related code from Emotisponge (which I keep hoping to get back to).

Source:

https://github.com/hartsick/itstwentyfifteen

[52bots]

#6 Winterscapes

What:

Generates wintry unicode landscapes 4x per day.

How:

Uses a selection of unicode characters for ground (trees, houses, horses, people) and sky (stars, snowflakes) with varied probabilities of appearance to generate a landscape with one row of ground, x of sky, and at least one snowman, and then tweets it.

So humans may actually follow the account, only tweets about every 3 hours between 7am and 7pm PST (~4x per day). Inspired by @tiny_star_field by @katierosepipkin. 

Why:

Winter! Also, Ascii art / animation experimentation. 

Source:

https://github.com/hartsick/winterscapes

[52bots]

#5 Slothrobot

What:

Pulls and tweets 140 characters from Thomas Pynchon's Gravity's Rainbow once per hour.

How:

Parses and removes all whitespace from the text, storing it in redis. Pulls a random 140 characters out, removing it from the text, then tweets and stores the new text back in redis for next time.

Why:

Wanted to do a mildly interesting "every X" bot that ended at some point. At less than 100 lines of code, this took more time to deploy than create.

Source:

https://github.com/jimmytheleaf/slothrobot

[52bots]

#4 Emotisponge

What:

An emotive bot whose mood and survival depends on human attention.

How:

Using the Twitter streaming API, listens for any user interaction with the bot. Once event is received, bot determines type of event, increments the appropriate counter in redis, and saves an appropriate response (DM, fave, tweet, etc.) to a redis queue.

At a set interval below rate limit, a separate process sends queued responses via Twitter's REST API.

Periodically and on demand, the bot calculates its own emotional score by tallying the redis counters and weighing them, reports on its status, and, potentially, dies.

DISCLOSURE: still a work in progress -- lots o' bugs. Will announce when ready for full release at @_hartsick.

Why:

Tinker with the Twitter Streaming API before tackling bigger interactive bots, and begin experimenting with user/bot interaction.

Source:

https://github.com/hartsick/emotisponge

[52bots]

#3 Wordchallenger

What:

A bot that attempts to win the Thesaurus Challenge.

How:

Drawing on the same Roget's 1911 source material, attempt to reverse engineer the game to get the right answers. So as not to spoil the fun for everyone else, only make an attempt every hour or so during the hours of 8am-8pm.

Why:

Solving game felt like such a well-formulated technical problem that I couldn't resist trying to solve it once I started thinking about it. I like the idea of twitter games, and also the idea of a bot playing a game another bot makes.

Source:

https://github.com/jimmytheleaf/ThesaurusGame

[52bots]

#2 HwaetBot

What:

Once per hour, generates a riddle in the Anglo-Saxon poetic style.

How:

Drawing on the riddles found in the Exeter Book, uses a Markov model with bigrams to generate novel sentences until 140 character limit is reached. Checks ensure unique tweets and that contain no full sentences from the original text.

Why:

I'm interested in playing around with different NLP techniques, and have a soft spot for Old English.

Source:

https://github.com/jimmytheleaf/hwaetbot

[52bots]

#1 Rosenbot

What:

Pulls sentences of less than 140 characters from Philip K. Dick's Do Androids Dream of Electric Sheep and tweets them once per hour.

How:

Pulls in a text (in this case, DADoES) and uses Python's Natural Language Toolkit to break the corpus into sentences. Selects a random sentence under 140 tweets in length, checks redis store to see if it has been tweeted before, and tweets it if not. Tweet text is then stored in redis so as not to be tweeted again in future.

Why:

Because 1) easing into using Python, Redis, Twitter API, and natural language processing for the first time, 2) Bladerunner showing at Cinespia, and 3) ROBOTS!

Source:

https://github.com/hartsick/rosenbot