After a couple of years of cancelled bookings due to COVID-19, this year was the first time I’ve been able to attend AWS re:Invent. What a conference. Unlike any other event I’ve ever seen, the scale of this thing is wild, even for Vegas. Everyone warned me about travelling through LAX on the busiest travel day of the year. But overall, things were pretty smooth. After over 24 hours of flights and layovers, the trip from Adelaide->Sydney->Los Angeles->Las Vegas ended in a wonderful thanksgiving lunch with friends I hadn’t seen since 2019.

The conference was incredible. But here are five lessons for next time…

---

1. It’s pretty spread out

The conference is spread out over multiple properties, ranging from a five minute walk to a shuttle bus ride. Staying within walking distance of the exhibit hall makes it easy to attend the morning keynotes. I stayed at the Palazzo which was probably the nicest hotel I’ve stayed at in Vegas. Bonus: AWS provide good rates if you get in early and book when you purchase a conference pass.

2. It’s not a big deal if you can reserve a seat for a talk

There are plenty of walk-up seats available for every talk. When I didn’t have a reserved seat, I got there 30-60 minutes early and never missed out.

3. Not everything is in the name

We use AWS CoPilot heavily and were pretty disappointed that searching the program only turned up a couple of talks on the topic. But after going to a talk on ECS and seeing them use CoPilot the whole time, despite not including it in the description, we realised we may have missed a lot of valuable sessions!

4. re:Play is a party

re:Play is the big party on Thursday night. It’s not a repeat viewing of previous sessions, not a board game night, and definitely not a movie night. Probably take it easy on Wednesday night in anticipation of this.

5. Leave Friday afternoon or Saturday

The exhibit hall closes on Thursday afternoon, the big party is Thursday night, and the talks wind up by lunchtime on Friday. All of this makes a Friday or Saturday morning exit recommended.

• Introduction to Google Cloud Code released earlier this year at Google Cloud Next. Looks like an awesome way of getting the inner development loop up and running much faster
• Lots of talk of “shifting left” (i.e. shifting the things that normally happen later in the development cycle to earlier on).
• Luke Million seeing us out with a DJ set was pretty cool.

---

We’re using Kubernetes on GCP for our entire platform these days, so I really wanted to get along to a conference and immerse myself in it.

---

It was good, but (and I should have expected it in hindsight) far more enterprise focused than something like PyCon.

Still, there was a good track for in-the-weeds development and I learnt a lot about CLI tools for working with Kubernetes.

The conference has now finished and I’ve travelled to Moscow to visit SkyOptix contact lens lab. We’re looking at partnering with them to manufacture and distribute our lenses in Russia and parts of Europe.

This was a pretty cool conference, right on the lake. Topics-wise, Machine Learning was big this year, it seems to be more and more popular every year.

Overall, a great two weeks in Cleveland. I found a good CrossFit gym nearby so I could get in my lifts during the conference, and afterward caught the bus to NYC.

---

At the Melbourne Convention and Exhibition Centre (MCEC). Obviously, much smaller than PyCon US but still worth attending for a couple of days. No sprints etc.

---

Here are some notes from PyCon this year in Portland Oregon. My second time in Portland.

---

THE COBBLER’S CHILDREN HAVE NO SHOES, OR BUILDING BETTER TOOLS FOR OURSELVES

Talk by Alex Gaynor.

• Good talk on writing CLIs for internal use - turning functionality into scripts and using argparse to define the interface

LASER CUTTERS, 3D PRINTERS, AND PYTHON

• SVG files are just XML! Way simpler than I thought.
• Check out OpenSCAD software.

VISUAL DIAGNOSTICS FOR MORE INFORMED MACHINE LEARNING: WITHIN AND BEYOND SCIKIT-LEARN

• More about how good sci kit is.
• More general machine learning talk.
• Pandas for loading data. What is it?
• She thinks that the future of machine learning practitioners looks like her. No formal training. Because of Python and scikit learn, in part.
• Informed machine learning is hard.

BETTER TESTING WITH LESS CODE: PROPERTY BASED TESTING WITH PYTHON

• Testing frameworks. Can generate data for you, and you can massage it to make the data fit your design.
• Property based testing - hypothesis - it helps find bugs
• look at slides! Should have some good links.
• Can bump up the number of tests in travis. I.e have it run quickly locally
• Recommend breaking up - fast unit tests and property based testing separately

PROTOTYPING NEW APIS WITH FLASK

• Check out marshmallow for data validation.
• Also use flask_login for login. There’s another one that was recommended by that British guy. It provides login pages that you can override etc. not sure if this was flask_login or not.

BUILD SERVERLESS REALTIME DATA PIPELINES WITH PYTHON AND AWS LAMBDA

• AWS lambda, only pay for the time the function runs.
• Could be good for realtime calculations for EyeSpace 3
• Stateless, scalable.
• Looks like you need to upload by putting code and dependencies into a zip file. Could this work with numpy?
• Deployment: Can do AWS CLI from travis CI job. How does this fit in with the zip file above?
• What is API gateway?

WHEN IS IT GOOD TO BE BAD? WEB SCRAPING AND DATA ANALYSIS OF NHL PENALTIES

• Need to look at pandas!
• Beautiful soup.
• “The brother problem”. I like the idea of naming a problem.

STATISTICS FOR HACKERS

• Basically, this talk looks at how you can “simulate” the scenario to calculate p value.
• Shuffling approach works when the null hypothesis assumes two groups are equivalent.
• Bootstrap sampling kinda works by resanpling the data.
• Protect against overfitting by cross validation. It is the go-to method for machine learning where you can’t really do an analytic analysis.
• Direct simulation method used in astronomy. Crazy example of simulating the night sky!

Other notes

• GitHub issues becoming popular, probably should switch to it for our internal issue tracker
• Visual screenshots diffing for tests could be a good idea