In a recent conversation I mentioned that when I meet a new Python feature, I add it to my refactoring checklist. Someone asked me to share that checklist, and here it is, slightly expanded:

• write all comments/docstrings
• resolve all TODOs
• add types to all fn declarations and key local vars
• expand short names to something explanatory
• for loops -> list comprehensions
• use dict comprehensions
• use generator or lazy iterator?
• use collections.Counter to count things
• use map, filter, reduce?

Obviously this doesn't address the "main" work of refactoring, like moving chunks of code around. But it's a list of stuff to look out for before pushing for public consumption, to encourage my code to be elegant, concise, and useable.

In other news, I'm looking for excuses to use the Heap computing cluster, which means a machine learning problem with significant HD or processor requirements. So far I'm considering:

• restarting my Bengali.AI OCR project
• doing something involving text prediction/generation, like Byronic poetry or Shakespearean sonnets or murder ballads
  • definitely want a PD corpus
  • pairing with a colleague taught me about beam search, which generates several high-scoring text strings rather than simply the best token-by-token one
  • inspiration: text generation with an RNN, Coursera NLP with Tensorflow Week 4
• helping experiment with a colleague's image captioning AI
• some Kaggle competition TBD?