serhii.net

loginctl as a way to manage sessions of logged in users

Instead of killing all processes belonging to someone, loginctl will return all sessions, and loginctl kill-session $number will log the user off!

New non-work user account!

Set my old Lain background with feh. I should look at some of my old i3 settings etc, to make it look different from the work one.

• zsh theme: trapd00r
• vim theme: pablo

General plans for vacation

• Don’t touch any ‘should’s - java, python, …; mostly focus on ‘housekeeping’ things if I want to do stuff with the computer
• Learn to use kitty well
  • Highlighting/copying URIs etc especially
  • See if I can use it to replace some of my hacks
• Learn to use tmux or screen well
  • Screen is available almost everywhere, but tmux is ‘better’
• Learn to use vim much better
  • Make an effort to learn it systematically and well from the beginning, I have a lot of antipatterns
  • w/E etc
  • vim recovery of swap files
• Sort out my i3/user config and all configs in general
  • Something easy to enable/disable keyboards (xinput float ..)
  • Something to turn on/off audio/webcamera
  • Move container splitting keybinding further away from window closing keybinding
• Sort out all dotfiles
  • A place where they are by default and can be imported/overwritten
  • In general any kind of dotfiles management/backup
• Sort out my startup scripts
  • In general something that doesn’t make me afraid to disconnect the laptop from the screen
    • Automatically use connected screens, without arandr-ing every time
    • Same for keyboards
    • Same for keyboard layouts
    • Run redshift and stuff only once
    • Even just an i3 keybinding that sets up what’s needed

“Ricing” - English / Unix / …

• Ricing is “making improvements to a system that don’t actually do anyone any good, and can sometimes have negative ramifications” ¹
• “Rice” is “a word that is commonly used to refer to making visual improvements and customizations on one’s desktop. It was inherited from the practice of customizing cheap Asian import cars to make them appear to be faster than they actually were” ²

(was curious about the name of a PPA)

i3 stuff

• lpicanco/i3-autodisplay: i3wm multiple monitors auto configuration

Test config file:

displays:
  - name: eDP-1
    workspaces: [1, 0]
    randr_extra_options: "--primary --mode 2560x1440"
  - name: HDMI-2
    workspaces: [2, 3, 4]
    randr_extra_options: "--above eDP-1"

• A long time ago I wrote something like this, but worse: nonpop/xkblayout-state: A small command-line program to get/set the current keyboard layout..

autorandr for flexible multimonitor setup

This is even better than the above: phillipberndt/autorandr: Auto-detect the connected display hardware and load the appropriate X11 setup using xrandr It saves configs readably and automatically to ~/.config/autorandr/config

General small things

• autorandr set up
• i3lock
• better autostart
  • start the diensttagebuch, work notes in the correct workspaces
  • start slack, Telegram and co in the correct workspaces
  • Put workspaces on the correct screens

i3-gaps fun

Very simple config:

gaps inner 10
gaps outer 10

Installed compton to get transparent terminals. Added this to kitty config:

background_opacity 0.8

Git use use specific public key file

When using public key and ssh for git, when you can’t use ssh-add ..., this works: GIT_SSH_COMMAND="ssh -i ~/.ssh/id_rsa_example" git clone example ³

Pycharm / matplotlib / pyplot debugging

I can happily use plt.plot()/plt.imshow() inside the <Alt-F8> and debugger console windows, it’ll be shown!

Recursively change owner in files owned by other user in current directory

Replace -user root with source user, $USER expands to user currently running command:

sudo find ~ -type d -user root -exec sudo chown -R $USER: {} +

sshfs / ‘Transport endpoint not connected’

In line with Day 784 about unmounting broken endpoints, yesterday I got a lot of errors (thunar didn’t start, I blamed memory, but df -h also didn’t start…), at the end the issue was with a sshfs directory:

fuse: bad mount point ./mountpoint’: Transport endpoint is not connected`

Using day 784 didn’t help, still got the above error. This helped: fusermount -uz myserver

Also, TODO: Why doesn’t linking stuff like this work?

{%raw%}
[Day 784]({% post_url 2021-02-23-day784.markdown %})
{%endraw%}

numpy true booleans

a is True is false for a numpy array of one element a, even if it’s value is True. a == True works correctly. Why does this happen?

pycharm debugging in console

You can use the console not just to look for output, but to interact with the variables etc! Why didn’t I think of this before: Using Debug Console | PyCharm

OpenCV documentation

I like giving code examples in C++, Java and Python for the same help topic! OpenCV: Creating Bounding boxes and circles for contours

Disabling touchpad while typing (xinput)

(22:31:53/11773)~/$ xinput list-props 15
Device 'SynPS/2 Synaptics TouchPad':
	Device Enabled (170):	1
	Coordinate Transformation Matrix (172):	1.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, 0.000000, 0.000000, 1.000000
	Device Accel Profile (304):	1
	Device Accel Constant Deceleration (305):	2.500000
	Device Accel Adaptive Deceleration (306):	1.000000
	Device Accel Velocity Scaling (307):	12.500000
	Synaptics Edges (327):	1574, 5368, 1408, 4444
	Synaptics Finger (328):	25, 30, 0
	Synaptics Tap Time (329):	180
	Synaptics Tap Move (330):	248
	Synaptics Tap Durations (331):	180, 180, 100
	Synaptics ClickPad (332):	1
	Synaptics Middle Button Timeout (333):	0
	Synaptics Two-Finger Pressure (334):	282
	Synaptics Two-Finger Width (335):	7
	Synaptics Scrolling Distance (336):	112, 112
	Synaptics Edge Scrolling (337):	1, 0, 0
	Synaptics Two-Finger Scrolling (338):	1, 0
	Synaptics Move Speed (339):	1.000000, 1.750000, 0.035417, 0.000000
	Synaptics Off (340):	0
	Synaptics Locked Drags (341):	0
	Synaptics Locked Drags Timeout (342):	5000
	Synaptics Tap Action (343):	2, 3, 0, 0, 1, 3, 0
	Synaptics Click Action (344):	1, 3, 0
	Synaptics Circular Scrolling (345):	0
	Synaptics Circular Scrolling Distance (346):	0.100000
	Synaptics Circular Scrolling Trigger (347):	0
	Synaptics Circular Pad (348):	0
	Synaptics Palm Detection (349):	0
	Synaptics Palm Dimensions (350):	10, 200
	Synaptics Coasting Speed (351):	20.000000, 50.000000
	Synaptics Pressure Motion (352):	30, 160
	Synaptics Pressure Motion Factor (353):	1.000000, 1.000000
	Synaptics Resolution Detect (354):	1
	Synaptics Grab Event Device (355):	0
	Synaptics Gestures (356):	1
	Synaptics Capabilities (357):	1, 0, 0, 1, 1, 1, 1
	Synaptics Pad Resolution (358):	54, 45
	Synaptics Area (359):	0, 0, 0, 0
	Synaptics Soft Button Areas (360):	3471, 0, 4054, 0, 0, 0, 0, 0
	Synaptics Noise Cancellation (361):	28, 28
	Device Product ID (297):	2, 7
	Device Node (296):	"/dev/input/event5"
(22:31:59/11774)~/$ xinput set-prop 15 349 1

Google Hangouts highlighting people

If there are too many people with video on, Google Hangouts moves the ones who talk closer to the beginning, making them visible?

pycharm/intellij running config environment variables spaces

Got bitten yet again when copypasting them - the name of one of them had four leading tabs. THAT DIDN"T GET SHOWN UNTIL I TRIED TO EDIT THE ENVIRONMENT VARIABLE IN THE PYCHARM WINDOW - it removes them when visualizing. Why? (The parameter of the last one had a trailing space too)

Python negative 0

-0.0 exists as float, and gets stored like this. Though it’s not less than 0 or +0.0. Can’t easily google a way to detect if it’s a negative 0 or not.

Signature detection

• By Microsoft: Making sense of Handwritten Sections in Scanned Documents using the Azure ML Package for Computer Vision and Azure Cognitive Services | CSE Developer Blog

• Aesthetically very pleasant category: Category:SVG signatures of musicians - Wikimedia Commons

  • Why did no one ever make a dataset out of this?
  • How to download all image files in a Wikimedia Commons page or directory | How To Wiki | Fandom
  • A lot of tools for dealing with SVG files in python
  • One could get a signatrue, distort it in all possible ways, add noise to the location/parameters of the points/curves
  • Specifically for paths: svgpathtools · PyPI

Random / CLI / CLI task manager / replacement for screen/tmux

GitHub - Nukesor/pueue: Manage your shell commands. (thank you AA)

Naming cheatsheet

GitHub - kettanaito/naming-cheatsheet: Comprehensive language-agnostic guidelines on variables naming. Home of the A/HC/LC pattern. (thank you AA)

From it:

Detectron2 dataloader training in parallel num_workers (“process exited unexpectedly”)

When training on different GPUs on the same server, I get errors like RuntimeError: DataLoader worker (pid 30141) exited unexpectedly with exit code 1.

The fix was to set the number of workers to 0: ¹

cfg.DATALOADER.NUM_WORKERS = 2

Object detection / segmentation metrics & evaluation

• Nice description of the official COCO metrics: COCO - Common Objects in Context
• Extremely interesting: Diagnosing Error in Object Detectors, University of Illinois at Urbana-Champaign (linked above in “analysis code”)

From SO: ¹

[..]the only difference between mAP for object detection and instance segmentation is that when calculating overlaps between predictions and ground truths, one uses the pixel-wise IOU rather than bounding box IOU.

ROC curve / cutoff point

Finding an optimal cutoff point in a ROC curve is largely arbitrary (or ‘depending on what you need’ based on the actual thing). A lot of ways to find this. (Nice list here, but I’d see if I can find a paper with a good overview: data visualization - How to determine best cutoff point and its confidence interval using ROC curve in R? - Cross Validated)

Detectron2 internals

Nice series of posts on how Detectron2 works inside: Digging into Detectron 2 — part 1 | by Hiroto Honda | Medium

Paper with object detection metrics comparison with the focus on COCO & open source

Electronics | Free Full-Text | A Comparative Analysis of Object Detection Metrics with a Companion Open-Source Toolkit

Untderstanding model performance by looking at examples it got wrong but was confident about

From How to work with object detection datasets in COCO format | by Eric Hofesmann | Feb, 2021 | Towards Data Science:

The best way to build intuition about how your model performs is by looking at predictions that it was confident about but got wrong. With FiftyOne, this is easy. For example, let’s create a view into our dataset looking at the samples with the most false positives

More examples of the same: IoU a better detection evaluation metric | by Eric Hofesmann | Towards Data Science