Mike Lee Williams mikepqr

I wrote this down after I responded to a page today (a holiday) because it would've been a decent pairing opportunity for a couple of new people on my team. Second best is that people can read what I did afterwards and ask me any questions. And then I realized that there's nothing PagerDuty-specific or confidential in here, so I may as well share it wider. It's hardly an epic incident, but it's a good example of "doing the work", I think. I borrowed the "write down what you learned" approach from Julia "b0rk" Evans. It's a fantastic practice.

The PagerDuty incident: "Disk will be full in 12 hours. device:/dev/nvme0n1p1, host:stg-nomadusw2-client-..."

(Note for non-PD readers: We run Nomad where others might run Kubernetes.)

Here's the process I went through.

Noticed that the usual docker system prune -a -f didn't resolve it
Tried docker system prune -a -f and it cleared up 0B

NIPS ML in the Law symposium 2016 notes

Including notes for the second session and the first panel.

My three takeaways

Tech folks have a tendency to "fly in and fix everything." That feels like a dangerous approach here. It's far better to stand on the shoulders of existing legal precedent, which has studied fairness, discrimination, and bias for decades, even if that slows down progress.
Machine learning systems mirror and amplify bias by default. We cannot simply ignore sensitive attributes because the system averages loss over the majority. (Disparate mistreatment). Pithy corollary: this problem will only go away if we devote resources into making it go away.
Providing explanations for decisions is the only humane way to build automatic classification systems. Why? If I can't test a result, I can't contest it. If the decisions must be testable and explainable, they will be much more reliable as a result.

Aaron Roth: Quantitative tradeoffs between fairness and accuracy in machine learning

Local (OS X) Side

`~/Library/LaunchAgents/pbcopy.plist`

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
     <key>Label</key>
     <string>localhost.pbcopy</string>

	# Reconstructed via infocmp from file: /usr/share/terminfo/t/tmux-256color
	tmux-256color\|tmux with 256 colors,
	OTbs, OTpt, am, hs, km, mir, msgr, xenl, AX, G0,
	colors#256, cols#80, it#8, lines#24, pairs#32767, U8#1,
	acsc=++\,\,--..00``aaffgghhiijjkkllmmnnooppqqrrssttuuvvwwxxyyzz{{\|\|}}~~,
	bel=^G, blink=\E[5m, bold=\E[1m, cbt=\E[Z, civis=\E[?25l,
	clear=\E[H\E[J, cnorm=\E[34h\E[?25h, cr=^M,
	csr=\E[%i%p1%d;%p2%dr, cub=\E[%p1%dD, cub1=^H,
	cud=\E[%p1%dB, cud1=^J, cuf=\E[%p1%dC, cuf1=\E[C,
	cup=\E[%i%p1%d;%p2%dH, cuu=\E[%p1%dA, cuu1=\EM,

	#!/bin/sh

	# search mount point based on the device name
	search_devicename () {
	diskutil list -plist \| \
	plutil -convert json -r -o - - \| \
	jq -c '[.AllDisksAndPartitions[].APFSVolumes, .AllDisksAndPartitions[].Partitions] \| flatten \| map(select(has("VolumeName")))[] \| {volume: .VolumeName?, id: .DeviceIdentifier?}' \| \
	jq -r --arg volname "$1" 'select(.volume == $volname) \| .id?'
	}

	from collections import defaultdict

	def node():
	return defaultdict(node)

	def word_exists(word, node):
	if not word:
	return None in node
	return word_exists(word[1:], node[word[0]])

	"""
	Minimal character-level Vanilla RNN model. Written by Andrej Karpathy (@karpathy)
	BSD License
	"""
	import numpy as np

	# data I/O
	data = open('input.txt', 'r').read() # should be simple plain text file
	chars = list(set(data))
	data_size, vocab_size = len(data), len(chars)

	# You don't need Fog in Ruby or some other library to upload to S3 -- shell works perfectly fine
	# This is how I upload my new Sol Trader builds (http://soltrader.net)
	# Based on a modified script from here: http://tmont.com/blargh/2014/1/uploading-to-s3-in-bash

	S3KEY="my aws key"
	S3SECRET="my aws secret" # pass these in

	function putS3
	{
	path=$1