A tool to help me optimize my typing by measuring how do I type.
Requires Linux, works by reading input devices directly.
I wanted to measure what how I type and optimize my layout, but I didn't like the idea of installing keyloggers: they'd log passwords! Thus, a keylogger isn't something I should run on my work laptop, and without it I'd miss out on lots of data.
N-gram keylogger doesn't log everything you type sequentially. Instead, it logs three things:
- key frequency
- bigram frequency
- trigram frequency
and only saves the db to disk once in a while, so even comparing two consecutive states of the database shouldn't help the attacker much in figuring out what exactly did you type. Good passwords contain rare n-grams though, so it...
One can configure it to stop logging into windows with certain window titles
(i3 or `sway' only),
and/or suspend logging altogether when processes with certain names are running
(rescans are throttled though, so one must take a pause before typing
in order not to outrun the process detection).
It aims to log not keypresses (who cares?), but rather semantic stuff like
Control-Alt-a, % and workspace-switch 1.
That aspect needs configuration and improvements though.
N-grams don't matter if you take breaks.
Resting, typing hi, resting again, typing you and resting again
will leave you with the following trigrams:
... ... h
... h i
h i ...
i ... ...
... ... y
... y o
y o u
o u ...
u ... ...
This way your n-grams don't work across your lunch breaks and you'll be able to identify your favourite opening and finishing sequences.
Can be configured to attach a different 'context' string to every bit it logs. Could be days of the week, could be applications you use or even editor modes, it's up to you. Contexts with too few actions will get their data discarded though for security reasons, so the more different contexts you have, the longer your collection periods should be.
# ngram-keylogger collect /dev/input/by-path/platform-i8042-serio-0-event-kbd --config example/config/t184256.py
... run it for a while
Now let's analyze how efficient my layout is.
A colon (:) is typed with Shift+; and is thus harder to type.
Does this make sense?
# ngram-keylogger query keypresses ':,;' --count
3735 | :
52 | ;
Not at all, I should probably swap them around.
My underscore is nearly the same combo as my Enter (only the timing differs), because I type a lot of underscores, right?
# ngram-keylogger query keypresses 'enter,literal-_'
4.639339% | enter
0.162412% | _
# ngram-keylogger query keypresses 'enter,literal-_' --renormalize
96.617647% | enter
3.382353% | _
Wrong. I clearly don't.
Do I at least type the letters according to the English letter frequency (ETAION...)?
# ngram-keylogger query --limit=5 keypresses '[a-z]'
4.095611% | t
4.042651% | e
3.964975% | i
3.424778% | o
3.216467% | n
OK, close enough.
What about capital letters?
# ngram-keylogger query --limit=5 keypresses '[A-Z]'
1.062741% | S
0.430745% | L
0.353070% | C
0.338947% | A
0.314232% | T
Huh? How come every 100th action I type is a capital 'S'? What do I type it after?
# ngram-keylogger query --limit=5 bigrams '*' 'S' --renormalize
40.199336% | S | S
18.272425% | L | S
7.641196% | space | S
6.976744% | o | S
5.980066% | ... | S
Wonderful, I spam it repeatedly, type LS or oS. Super natural.
... S and space S at least make some sense.
So, where do I spam SS?
# ngram-keylogger query --limit 3 --by-context bigrams 'S' 'S' --renormalize
90.082645% | term:vi:magit:nrm | S | S
4.958678% | term:vi:magit:ins | S | S
3.305785% | browser | S | S
Whew, I'm not going crazy. S is a stage hunk shortcut in vimagit,
and I definitely do spam it when I review what I'm going to commit.
# ngram-keylogger query --limit 3 --by-context --contexts 'term:vi:magit*' keypresses --renormalize --cumulative
7.714808% | 7.714808% | term:vi:magit:nrm | down
13.747715% | 6.032907% | term:vi:magit:nrm | S
19.049360% | 5.301645% | term:vi:magit:nrm | up
Yeah, checks out. 1/5 of my vimagit keypresses are moving up and down and staging hunks.
Case closed.