PyData Recap

by bizzle

PyData ( is a fairly regular conference put on by NumFocus (, a non-profit that sponsors most of the big Python projects (NumPy, pandas, Jupyter, Julia, Bokeh…). Within the span of a few weeks, they’ve put on conferences in Washington DC, NYC and LA. Three weeks ago, I attended the PyData NYC version – here’s my highlights:

– Lead Developer for pandas (Jeff Reback) talked about the next version of pandas (due mid-November) which will treat NaN as a first class integer system (currently, NaN forces a column to convert to float64, which is really annoying when you’ve got Int64 columns).
– James Powell gave a really entertaining talk about the nitty-gritty of unicode and identifiers in Python (Poo Emoji is a valid identifier, but an ellipse is not – he dug into the C code to prove it). I had previously seen his “So you want to be a Python expert?” talk (, he’s fantastic.
– giphy is using session browse data as sentences, passing it into word2vec, and using that for gif recommendations. Sound familiar?
– Bokeh is worth checking out, Luke Canavan (developer of Bokeh) made a browser-side face recognition model easily
– Lightning Talks are amazing. Five minute presentations, hard cut-offs. It was really refreshing to see short subject presentations boiled down to the bare bones.
– Julia is getting hot.

So – should you attend one yourself? It was certainly not very academic: very few talks had papers associated with them. PyData struck me as much more business/hacking/results oriented. A general theme was, “hey, this works, isn’t this cool?” On the plus side, attending this conference very seriously supports open source (most presentations were by open source package authors) and it was inspiring to see short subject talks.

Bottom line: Soft Recommend