author: niplav, created: 2022-07-07, modified: 2024-07-08, language: english, status: maintenance, importance: 2, confidence: log
Notes for myself on the data I track, how to transform it into a usable shape, data quality and other random assortments.
I've always collected some data about myself and the world around me, but not using/analyzing it because of a chronic "I'll get around to it eventually" syndrome. Which is a shame, because that means I've been putting in a reasonably large amount of effort and have nothing to show for it, a 1-legged stool:
The QS cycle is straightforward and flexible:
- Have an idea
- Gather data
- Test the data
- Make a change; GOTO 1
Any of these steps can overlap: you may be collecting sleep data long before you have the idea (in the expectation that you will have an idea), or you may be making the change as part of the data in an experimental design, or you may inadvertently engage in a “natural experiment” before wondering what the effects were (perhaps the baby wakes you up on random nights and lets you infer the costs of poor sleep).
The point is not publishable scientific rigor. If you are the sort of person who wants to run such rigorous self-experiments, fantastic! The point is making your life better, for which scientific certainty is not necessary: imagine you are choosing between equally priced sleep pills and equal safety; the first sleep pill will make you go to sleep faster by 1 minute and has been validated in countless scientific trials, and while the second sleep pill has in the past week has ended the sweaty nightmares that have plagued you every few days since childhood but alas has only a few small trials in its favor—which would you choose? I would choose the second pill! […]
One failure mode which is particularly dangerous for QSers is to overdo the data collection and collect masses of data they never use. Famous computer entrepreneur & mathematician Stephen Wolfram exemplified this for me in March 2012 with his lengthy blog post “The Personal Analytics of My Life” in which he did some impressive graphing and exploration of data from 1989 to 2012: a third of a million (!) emails, full keyboard logging, calendar, phone call logs (with missed calls include), a pedometer, revision history of his tome A New Kind of Science, file types accessed per date, parsing scanned documents for dates, a treadmill, and perhaps more he didn’t mention. […]
One thinks of a saying of W. Edwards Deming: “Experience by itself teaches nothing.” Indeed. A QS experiment is a 4-legged beast: if any leg is far too short or far too long, it can’t carry our burdens.
—Gwern, “Zeo sleep self-experiments”, 2018
At least I now know that I'm falling into this trap, "Selbsterkenntnis ist der erste Schritt zur Besserung". And the second step is to bring all of your data in a usable format.
I use spaced repetition, and plan to take it as a proxy for cognitive performance in QS experiments.
The data can be found in the helpfully named collection.anki2, which
is actually an sqlite database
in disguise.
The Anki manual
helpfully informs that the most important table is revlog, one can
then export the data to CSV with the following command:
echo -e '.headers on \n select * from revlog;' |
sqlite3 anki_2022-07-04T08:43:00.db |
tr '|' ',' >anki_2022-07-04T08:43:00.csv
The header is to be interpreted as follows:
The most important table for statistics is the revlog table, which stores an entry for each review that you conduct. The columns are as follows:
id
The time at which the review was conducted, as the number of milliseconds that had passed since midnight UTC on January 1, 1970. (This is sometimes known as Unix epoch time, especially when in straight seconds instead of milliseconds.)
cid
The ID of the card that was reviewed. You can look up this value in the id field of the cards table to get more information about the card, although note that the card could have changed between when the revlog entry was recorded and when you are looking it up. It is also the millisecond timestamp of the card’s creation time.
usn
This column is used to keep track of the sync state of reviews and provides no useful information for analysis.
ease
Which button you pressed at the end of the review (1 for Again, 4 for Easy).
ivl
The new interval that the card was pushed to after the review. Positive values are in days; negative values are in seconds (for learning cards).
lastIvl
The interval the card had before the review. Cards introduced for the first time have a last interval equal to the Again delay.
factor
The new ease factor of the card in permille (parts per thousand). If the ease factor is 2500, the card’s interval will be multiplied by 2.5 the next time you press Good.
time
The amount of time (in milliseconds) you spent on the question and answer sides of the card before selecting an ease button.
type
This is 0 for learning cards, 1 for review cards, 2 for relearn cards, and 3 for "cram" cards (cards being studied in a filtered deck when they are not due).
— Anki developers, “Manual Analysis” in “Graphs and Statistics”, year unknown
The CSV of the data can be found here.
Similarly, one can export meditation data from Medativo (if one has coughed up 5€ for the premium version, which I decided was worth it for the data, after having locked myself in :-|):
echo -e '.headers on \n select * from History;' |
sqlite3 meditation_2022-07-02T20:00:00.db |
tr '|' ',' >meditations.csv
The names for the columns are exceedingly obvious and need no further explanation.
I didn't rate my sessions in the beginning (and manually inserted data from meditation retreats with unrated sessions), leading to a very optimistic default of 4.0 mindfulness and "concentration" (better called absorption, I claim). So we remove those, using the sam language in vis:
,/^1,/;/^860,/
x/4\.0,4\.0,/c/,/
,/^1210,/;/^1308,/
x/4\.0,4\.0,/c/,/
,/^1594,/;/^1615,/
x/4\.0,4\.0,/c/,/
The CSV of the meditation data can be found here.
mindfulness_ranking and concentration_ranking are both subjective
impressions directly after meditation, where "mindfulness" describes the
degree of sensory clarity, and "concentration" (better called "absorption"
or "rest") describes my ability to rest on a specific sensory object.
Sanitizing the sessions file by converting the datetime to ISO-8601 using structural regular expressions, and some other minor fixes:
,x/([0-9]+)\/([0-9]+)\/([0-9]+) /c/\3-\1-\2T/
,x/; /c/;/
,x/(T[0-9]+:[0-9]+),/c/\1:00,/
,x/-([0-9])-/c/-0\1-/
Formatting the approaches file:
,x/ ,/c/,/
Find incorrectly written locations:
$ awk -F, '{ print($2) }' <daygame_approaches.csv | sort | uniq
and manually correct them (this is useful for the other fields as well, just to check consistency).
Anonymizing locations and the names of the women:
$ awk -F, 'BEGIN {
FS = OFS = ","
while (getline < "admn/daygame/locations") {
loc[$2] = $1
}
close("locations")
}
{
if (name[$8] == "" && $8 != "Name") {
name[$8] = int(100000 * rand())
gsub(/\./, "", name[$8])
}
if ($2 != "Location") {
original_location = $2
gsub(/"/, "", original_location)
if (loc[original_location] == "") {
print "Warning: Location '" $2 "' not found in locations file" > "/dev/stderr"
loc[original_location] = 100000 * rand()
printf "%d,\"%s\"\n", id, location >> "admn/daygame/locations"
}
$2 = loc[original_location]
}
if ($8 != "Name") {
$8 = name[$8]
}
print $0
}' <daygame_approaches.csv >daygame_approaches_anon.csv
$ mv daygame_approaches.csv daygame_approaches_deanon.csv
$ mv daygame_approaches_anon.csv daygame_approaches.csv
The approaches file can be found here, the sessions file can be found here.
Approaches file datapoints (in CSV):
Sessions file:
I use the Fitbit Inspire 3 because Fitbit is one of the few (the only?) companies whose products allow for data exporting, mostly to track my sleep, but maybe I'll also get mileage out of the heart rate, step, glucose and temperature tracking.
Sleep data here.
Other metrics I track don't deserve as much elaboration.
I track when I masturbate & how good it feels & the type of pornography in this file via this script. Data quality is pretty high.
t stands for texta stands for audioi stands for imagev stands for videoI track my mood via the excellent Mood Patterns which performs experience sampling allows swift CSV export of the data. They even turned changed the annoying "hitting a block of wood with a hammer" notification sound to the OS default. No post-processing needed, the data is just there. An app by programmers, for programmers.
But there is still some data cleanup to do:
sort -n mood.csv >>~/proj/site/data/mood.csv
Finally I rename the mood columns simply to "happy", "content", "relaxed", and "horny".
CSV here, the data quality is mediocre (long stretches of not responding to questions, giving more conservative (closer to 50) answers over time, starting to use activities around July 2022 (and not using it for what it was intended for: when the activity is "Nothing" that means I carried on with my day as normal afterwards, if the activity is "Mindfulness" it means I spend a couple of seconds in a more mindful state)). Also, I use the "interested — uninterested" metric to track horniness (higher means hornier).
I track which substances I take (nootropics/melatonin/drugs) in this file via this script. Data quality is good but fairly few entries. At the moment I am mostly using it to perform self-blinded RCTs.
Tracking weight, mostly for exercise purposes. Data here, collected with this script.
Productivity, creativity and the subjective length of the day. Collected with this script into this file. Started collecting subjective length of the day on 2023-08-21.
Data on bag spreading on public transport, in this file. Data quality is horrible: probably prone to multiple biases from my side, from different locations, no tracking of location or datetime…maybe I should just delete this one.
Via Sensor Logger, I use my phone as an easy way to collect large amounts of data. I don't want to make the files public, as they contain information that could de-pseudonymise me.
In principle it should be possible for me to track my forecasting performance on Manifold, Fatebook/PredictionBook and Metaculus, given that all of them have APIs over which data can be exported and analyzed. In practice I haven't done so yet, but it might be a good (albeit slow-to-evaluate) proxy for cognitive performance.