Does use of Twitter discourage exploring ideas in greater depth?

Twitter makes it incredibly easy to share random thoughts and links to topics of interest. For me personally, this is the main attraction of using Twitter, but the drawback of the brevity is that it discourages exploring and expanding a thought beyond 280 characters (how did we ever survive with only 140 characters?!)

After posting a short sentence about something I’m thinking about or have recently read elsewhere, I usually think “I could write a whole article on this single topic”. For example:

I could have easily taken this idea of open office spaces having the opposite intended effect on worker productivity and explored this in more depth in a longer article, but instead captured this thought as a single paragraph and left it at that.

In this sense I think Twitter makes us lazy. It makes it easy to quickly share a quick thought, but in doing so we throw out these nuggets of info and then leave them there, unexplored.

I don’t make New Years Resolutions, but if there’s one thing I plan to do more of this year, it’s to spend more time writing more articles, and less time sharing quick, throwaway thoughts.

Building bots on Twitter with AWS Lambdas (and other stuff)

I’ve built a few different bots on Twitter and written several articles describing how I built them. Some of these were a few months back – once they’re up and running it’s easy to forget they’re up and running (thanks to the free tier on AWS Lambda which means you can run scheduled Tweets well within the free tier limits). This is a summary of the bots I’ve developed so far.

Looking at where I got started, my first bot was to build an integration between Amateur Radio 2m Packet, retweeting packets received locally to Twitter. This was my first experience working with the Twitter REST apis and the OAUTH authentication, so I lot of what I learned here I reapplied to the following bots too:

For my next project, I was inspired by articles by researcher Janelle Shane who has been training ML models to produce some hilarious results, such as weird recipes, college course names and many others. I was curious what content a ML model would generate if I extracted all of my past 4000+ Tweets from Twitter and trained a model with the content. I had many questions, such as would the content be similar in style, and is 4000 Tweets enough text to train a model? You can follow my progress in these posts:

This then led to repeating the experiment with over 10 years of my blog articles and posts collected here, which you can follow in these posts:

Next, what would it take to train my model in the cloud using AWS Sagemaker, and run using AWS Lambdas?

You can follow this bot on Twitter here: @kevinhookebot

I had fun developing @kevinhookebot – it evolved over time to support a few features, not just to retweet content from the trained ML model. Additional features added:

  • an additional Lambda that consumes the Twitter API ‘mentions’ timeline and replies with one of a number of canned responses (not generated, they’re just hard coded phrases). If you reply to any of it’s tweets or Tweet @ the bot it will reply to you every 5 minutes when it sees a new tweet in the mentions timeline
  • another Lambda that responds to @ mentions to the bot as if it is a text-base adventure game. Tweet ‘@kevinhookebot go north’ (or east/west/south) and the bot will respond with some generated text in the style of an adventure game. There’s no actual game to play and it doesn’t track your state, but each response is generated using @GalaxyKate ‘s Tracery library to generate the text using a simple grammar that defines the structure of each reply.

After having fun with the adventure text reply generator, I also used the Tracey library for another AWS Lambda bot that generates product/project names and tweets every 6 hours. I think it’s rather amusing, you can check it out here: @ProductNameBot

@ProductNameBot

My most recent creation I upped the ante slightly and wondered what it would take to develop a Twitter bot that playeda card game. This introduced some interesting problems that I hadn’t thought about yet, like how to track the game state for each player. I captured the development in these posts here:

I have some other ideas for something I might put together soon. Stay posted for more details 🙂

Using AWS SageMaker to train a model to generate text (part 3): What I learned about Python AWS Lambdas this week

This is a follow-on from my investigation on how to use AWS SageMaker as an AWS replacement to my current approach to generate text from a Machine Learning model. Up until this point I’ve been running torch-rnn on a server locally. You can follow part 1 and part 2 of my progress so far.

In summary, here’s what I learned this week:

  • Some Python modules are OS platform specific. That means, if you install a module on MacOS, you can’t zip it up as a dependency in a .zip deployment for a Lambda which runs on Linux, as it won’t be OS compatible
  • The maximum size for an AWS Lambda deployment is 50MB. Zipping up what I’ve built so far (only a minimal script but relying on a number of modules) I’ve got a 500MB zip file. Clearly that’s too large to deploy as a Lambda
  • Following some suggestions here,  there are Python frameworks (such as Zappa) to help build Python based AWS Lambdas and address some of the issues with modules and deployment. Clearly I’ve got some learning here to get his to work 🙂