In the middle of the desert you can say anything you want
I can’t believe I have to say this, but please don’t drink bleach.
— Joe Biden (@JoeBiden) April 24, 2020
We live in interesting times.
shambolic - Chaotic, disorganised or mismanaged. Found somewhere on The Guardian applied to Trump’s daily health briefings.
A ! either forces the action or toggles the action:
:set cursorline, to turn off: :set nocursorline
Is equivalent to:
:set cursorline! 1
tabulate2 generates nice tables of various formats! Things like:
print(tabulate.tabulate(db,headers=db.columns))
epoch loss val_loss val f-score
-- ------- ------ ---------- -------------
0 1 4.31 4.62 0.579
1 2 3.72 3.61 0.705
2 3 3.54 3.25 0.722
3 4 3.31 3.06 0.737
4 5 3.19 2.93 0.736
5 1 4.31 4.62 0.581
6 2 3.72 3.61 0.72
7 3 3.54 3.25 0.755
8 4 3.31 3.06 0.755
9 5 3.19 2.93 0.764
10 6 3.12 2.83 0.798
11 7 2.95 2.76 0.779
12 8 2.91 2.69 0.757
13 9 2.84 2.64 0.816
14 10 2.68 2.63 0.835
15 11 2.71 2.56 0.83
16 12 2.69 2.52 0.825
17 13 2.62 2.49 0.826
18 14 2.6 2.46 0.845
19 15 2.56 2.44 0.84
tabulate · PyPI is the basic documentation with visualizations of each tablefmt. It even supports jira! And pipe is the usual markdown format. Let’s try:
| epoch | loss | val_loss | val f-score | |
|---|---|---|---|---|
| 0 | 1 | 4.31 | 4.62 | 0.579 |
| 1 | 2 | 3.72 | 3.61 | 0.705 |
| 2 | 3 | 3.54 | 3.25 | 0.722 |
| 3 | 4 | 3.31 | 3.06 | 0.737 |
| 4 | 5 | 3.19 | 2.93 | 0.736 |
| 5 | 1 | 4.31 | 4.62 | 0.581 |
| 6 | 2 | 3.72 | 3.61 | 0.72 |
| 7 | 3 | 3.54 | 3.25 | 0.755 |
| 8 | 4 | 3.31 | 3.06 | 0.755 |
| 9 | 5 | 3.19 | 2.93 | 0.764 |
| 10 | 6 | 3.12 | 2.83 | 0.798 |
| 11 | 7 | 2.95 | 2.76 | 0.779 |
| 12 | 8 | 2.91 | 2.69 | 0.757 |
| 13 | 9 | 2.84 | 2.64 | 0.816 |
| 14 | 10 | 2.68 | 2.63 | 0.835 |
| 15 | 11 | 2.71 | 2.56 | 0.83 |
| 16 | 12 | 2.69 | 2.52 | 0.825 |
| 17 | 13 | 2.62 | 2.49 | 0.826 |
| 18 | 14 | 2.6 | 2.46 | 0.845 |
| 19 | 15 | 2.56 | 2.44 | 0.84 |
How does Tensorflow train stuff when loss is nan? It keeps doing something, accuracy changes, etc etc etc. - is the gradient calculated per batch as normal,
Einstein / Netzah “do your own thing”
Zinseszins - Wiktionary - compound interest. Beyond logic
Split string with multiple delimiters in Python - Stack Overflow has this really nice idea of doing str.replace(':', '_') and then str.split('_').
Series.str.split(delim) can do Series.str.split(delim, expand=True), that returns different columns instead of a list.
Also - if graphs don’t make sense, one way things have gone wrong is interpreting x and y as strings.
c.tabs.background = True in config opens new tabs in the background when middle-clicking.
Nurse blocking anti lockdown protests in Denver : PublicFreakout and the imgur album: Health care workers stand in the street in counter-protest to hundreds of people who gathered at the State Capitol to demand the stay-at-home order be lifted in Denver, Colo., on Sunday, April 19, 2020. Photos by Alyson McClaran - Album on Imgur. As someone said, I really imagine seeing this in history books.
To visualize logs of screen sessions, less -r filename (-R hates ^H progress bars but I search seems to work a bit better) works very well. Even for the (tf) progress bars almost. ansifilter exists but deals with progress bars much worse.
Empirically - for searching and scrolling the same font size / terminal size is very helpful.
Just pasting the Vim macro on Day 469, while it did not break Jekyll this time (why?), seems to have broken the RSS script at the beginning of the page.
This page contains the following errors:
error on line 19 at column 73: PCDATA invalid Char value 27
Copypasted the lines from terminal and back, but I need a much better way to save and ‘publish’ the macros. Found Vim Macro Trickz • Hillel Wayne when looking for solutions.
.. can be added in a bath, but only in a carrier oil, olive oil works.
“For a single bath, three to 12 drops of essential oil in a tablespoon (15 ml) of carrier oil is sufficient to create a very aromatic, therapeutic bath” 1
Can I export a tensorflow summary to CSV? - Stack Overflow – yes I can, by clicking “show data download links” on the top left of interface and selecting what I want to download below the plots.
machine learning - Why training loss is increased at the beginning of each epoch? - Stack Overflow - probably because the dataset is not random enough, and the network overfits on the structure of the dataset. Happens with buffer_size dramatically less then the size of the structured dataset.
Added config.bind('tw', 'set-cmd-text -s :open -w') to config, now tw is like a, but for windows instead of tabs.
Added this macro:
let @R = 'viw"oyi[^^[^[^[^[A]^[^[^[Go[^^[^[^[^["opA]: ^[^[^[^['
When the last word in the line is selected, it gets transformed immediately to a reference. I’ll use this vim cheatsheet as 1.
Vim cheatsheetVim cheatsheet
As a bonus, this uses a non-default vim register so whatever is in my clipboard stays there. (But it destroys whatever I had in "o). ↩︎
Mafia distributes food to Italy’s struggling residents
In recent days, the police in Naples have intensified their presence in the poorest quarters of the city, where men tied to the Camorra, the Neapolitan mafia, have organised home delivery of food parcels. Magistrates have already begun an investigation against a group of people who were questioned while distributing food to local residents.
The Science of Happiness | Greater Good is a podcast about happiness, found it while looking for The Happiness Lab which is also a podcast about happiness.
Screens named with screen -S myname and attached by screen -r myname - tab completion works with screen -r!
entry is the time the task was created. So:
task all sprint:$SPRINT status:pending entry.before:eow-3wk tags.not:s rc.report.all.columns:entry,end,tags,description rc.report.all.labels=Age,Done,Tags,Description rc.report.all.sort:end rc.verbose:label,blank is what I’ll use in SIOM.
I missed this all my life.
If more than 1 GPU are visible, tensorflow will take some memory on all of them, even if its actively using only one. Restricting visibility the usual way helps.
task all status:completed end.after:eow-2wk – I forgot you could do almost-math with the datetimes!
Editing the output, to sort it by project, divide different ones by newlines, and show the headers of the table but nothing else:
task all status:completed end.after:eow-2wk rc.report.all.columns:project,entry,end,tags,description rc.report.all.labels=Project,Created,Finished,Tags,Description rc.report.all.sort:project-/,end rc.verbose:label
This is a really nice video: How to Make Perfect Pizza | Gennaro Contaldo - YouTube
Using an userscript for this is overkill, but
#!/usr/bin/python3
import os
title = os.environ['QUTE_TITLE']
url = os.environ['QUTE_URL']
with open(os.environ['QUTE_FIFO'], 'w') as f:
f.write("yank inline \"[{} {}]\"".format(url, title))
Located at /home/shamotskyi/.local/share/qutebrowser/userscripts/yank_w.py.
In the config, config.bind('yw', 'spawn --userscript yank_w.py')
EDIT: replaced this with config.bind('yw', 'yank inline "[{url} {title}]"').
I have not been using it to the fullest.
List tasks completed today: task all status:completed end:today
This might be something I can copypaste into my pages or in my reports, such as my weekly sprint review! Adding it as a module to siom should be not too hard. Taskwarrior - Usage Examples contains more nice examples.
glob is not sorted!
glob.glob().sorted() though is.
kitty - the fast, featureful, GPU based terminal emulator — kitty 0.17.2 documentation:
ctrl+shift+s is paste from selection
ctrl+shift+u for unicode characters
sudo update-alternatives --config x-www-browser
Doesn’t seem to work for kitty, but in the config open_url_with qutebrowser works predictably.
gratuitous - Wiktionary - given freely, unearned; without reason, unjustified, unnecessary, uncalled for. Seen first in Behemoth by Peter Watts.
For the sake of consistency (to each other and especially to Google Calendar) now my sprint number and week number is the ISO week number:
export SPRINT=$(date +%-V)
Fixed this in:
tztime local {
format = "[%V] %a %Y-%m-%d %H:%M:%S"
}
Only place it’s left now is in my paper calendar, but I don’t use it too much these days.