Lightning Talks
Really messy!
1. Linting
Originally a C program by Stephen Johnson, 1978 paper.
AST, checks against rules.
Checks code more strictly than a compiler for syntax, consistency, style
Linters are lint-like programs for many languages.
Purpose
Less Debugging, Looking for logic loopholes, Carving out dead code, Reduce runtime errors, For big teams: Consistency, Interpreted languages - no compile time
Uses
Unused variables, functions, dead code, undeclared/unset vars, enforcefunction argument/parameter checking... etc.
Style Checking
Code looks fine but is inconsistent. You know the stuff, PEP8. Consistency.
Python
Flake8, PyLint, PyFlakes, Pep8
For editors
vim - flake8
emacs - ?
Custom lint rules
You can write a plugin that examines whatever can be checked by AST, etc.
Also you can write little rulesets that "lint" your code by rules
2. Zebras and Lasers - Jonas Neubert - Zymergen
Talk about barcode readers.
Writes software stuck in a box for factories.
The laser takes samples over time, so you have to do image processing.
You thing opencv = image processing. I think LIST COMPREHENSIONS.
kernel = lambda a,b,c: a if a==c!=b else b
smooth = ''.join([(a,b,c) for a,b,c in zip(data[0:], data[1:], data[2:])])
[str(round(len(list(v))/4.0)) for k,v in itertools.groupby(smooth)][1:-1]
Look up what you got on the wikipedia table of barcodes
3. 2FA, OTP, WTF - Making technical concepts accessible - Kelly
Who here works on a website where people sign up or login? Congrats you are a security engineer.
Get your family using password managers and 2FA.
Factors: Something you know, something you have, something you are
Use this stuff.
4. Gert, Scientist @ Nasa working on Kepler satellite
Kepler looks at the brightness of stars and if you look long enough the brightness dims for 6 hours once a year.
You can find a planet using 7-8 lines of python.
I am going to find a planet around a distant star live on stage.
python3, numpy, scipy.signal, matplotlib working in a jupyter notebook
Use a low pass filter to remove long term trend in star brightness data, this normalizes the year.
Next, a lomb-scargle periodogram or something is used to find periodic signal in the data.
The planet has a 0.84 day orbit, so the data was aggregated daily.
Python package: PyKE - come talk to us if you want to be a developer or intern at nasa.
5. Bot Cars! - John Ellis
Race course out in oakland once a month, the cars look like RC type cars.
Were are these events? Huge warehouse in oakland!!
Behavior cloning is what we do.
- Other techniques: You can do line following, something else...
- Data labeling is tough in machine learning. Having a data labeler system expidites that.
Toolchain
- Camera input device
- OpenCV, B4L2 capture, python3
- Adafruit BNO005 gyroscope, dualshock4+ds4drv controller, usb2 camera v4l2capture
- nvidia jetson tx2 + orbitty carrier
- pololu micromaester + libusb
- Cycloid: Traxxas Chassis
- Donkeycar: raspberry pi, 3dprinted superstruture, ~$200
- Derp Learning: Nvidia TX1, Traxxas Slash, ~$800
- put a gpu on your car so you have no delay and can learn on board
- Virtual test environment - use a graphics engine in unity
- you can use numpy+scikit image but it has its limits, he did it
6. Quilt - Docker, but for Data - Anish
akarve/pydatabook - an example
Why?
- Code has reusable building blocks, github, PyPI, DockerHub
- Data is missing reusable building blocks - stuff like s3, github, everywhere online
- Data packages: metadata and serialized
Package Lifecycle - 3-4 primitives that matter
- build, , install, import
- `quilt build` - shred them into a dataframe, hash, stick into a hash tree, serialize columnar data, indexing
Why track in a hash tree? - Need to be able to send all the deltas
If you just change a readme, just want to download that on a version update
Catalog is like PyPI but for data
from quilt.data.akarve import pydata_book as pb - pb.getdata.train() (or something) and you get a dataframe out.
Working on react+Tensorflow, R, Spark, going to get stuff into Jupyter.
7. Datasette - Simon @ Eventbrite - journalist stuff @ guardian
Make an instant, read-only JSON API for any SQLite database. It also provides tools for packaging the database as a docker container and deploying to hosting providers.
Arbitrary SQL is pretty fine, you can't really modify the data, read only. Chrome history example.
San Francisco open data portal, data.sfgov.org - download csv of trees from the city
Wrote [csvs2sqlite] which just turns the CSV to a sqlite3 database. Can refactor csv to something a bit more relational. Extract some columns... He list about 12 columns he wants into separate tables. Also he uses -x to make the columns searchable in sqlite, indexing or whatever?
$ datasette tree.db
$ datasette publish now sf-trees.db - gives you a URL, hyperlinks.
Just like that you have a browseable, searchable json API
he built https://sf-tree-search.now.sh because he thought "why not?" - search sf by tree
8. How to build community - Simeon
Pacman rule - Eric Holscher - don't stand in a circle, people can't get in. Everyone should meet more people.
Bring a friend, bring a female coworker.
BayPiggies, SF Django Meetup, Pyninsula, PyLadies, East Bay Python, North Bay Python