[split] Anki's usefulness for QS

[quote]What are you learning? Learn it with an SRS & record your progress.[/quote]I really like Anki. I recommend it to every student who struggles with learning.

When it comes to using Anki scores for QS purposes the data however isn’t easy to interpret. I think Nick Winter did try to invest some effort into analysing the data and found it too noisy.
I myself have over a year of Anki data and didn’t learn anything specific from the data expect the fact that I get faster over time at solving Anki cards.

Don’t you mean, every students? :smiley:

You really can. In fact, if you want to lose a couple of kilograms quite quickly, you can just adjust your caloric intake to 0 for a day or so. I’m not necessarily recommending this strategy (although I do use intermittent fasting myself) but it is a quick & easy proof.

The claim you might have heard is that dieting doesn’t lead to persistent weight loss (ie, in the clinical trials, after the diet period is over, the subjects who lost weight gain it back). These trials are typically in morbidly obese patients – I have a feeling there’s something else going on with them that isn’t present in most of us (maybe something genetic, high baseline levels of inflammation, maybe something else).

There was a study I saw recently that showed college students who tracked their weight managed to avoid the typical weight gain associated with starting school (“the freshman 15” we called it at my university).

I knew this, but I acknowledged the mistake I made in simplifying my explanation.

I am told it’s probably genetic.

Probably because they choose what to eat more carefully.

Anyway, my current hypothesis is that carbohydrate is uniquely fattening, and is also associated with spiking blood sugar, giving me metabolic syndrome. This fact, I know, from simply diligently measuring myself everyday.

What do you mean by noisy? Jeremy Howard & Seth Roberts both found that walking on a treadmill enhanced their studying. I noticed the same thing.

I’m watching my stats to see what relationships there might be between diet, exercise, & learning. This week I had an exceptionally high score & I have a couple of theories about what might have provoked it.

At the very least, it is exciting to try to find relationships like this.

1 Like

It would be great if you link to those authors.

No problem.

Here is Seth Robert’s original post: http://blog.sethroberts.net/2010/02/23/boring-boring-pleasant/

He has a bunch of posts tagged “walking and learning”.

Jeremy Howard’s video is here: Jeremy Howard on Language Acquisition Performance - Quantified Self

I don’t know what other material he has out there, so if you come across anything, please share it with me.

[quote]I’m watching my stats to see what relationships there might be between diet, exercise, & learning. This week I had an exceptionally high score & I have a couple of theories about what might have provoked it.
[/quote]Which scores do you mean specifically?

[quote]What do you mean by noisy? Jeremy Howard & Seth Roberts both found that walking on a treadmill enhanced their studying. I noticed the same thing.
[/quote]To measure The effect of his general cognitive function Seth Roberts still use the math test. That suggests that he thinks the math test provides clearer data on the effect of something like eating a lot of butter.

There are a bunch of reasons why Anki’s default data is noisy. The difficulty of cards can vary from day to day. I’m not succeeding in doing all Anki cards every day. That means that the cards I will do the next day are harder than they should be. Sometimes you add a bunch of new cards that are really hard. Sometimes you add a bunch of new cards that are easy.

When Anki lists time/card it lists averages. Averages are bad. The median is much better.

Forunately Anki saves the raw data. It should be possible to get a better metric out of that raw data.

If someone wants to write a bachlor/master thesis is psychometrics this is a perfect topic. The data set of Mnemosyne is open. If someone does the necessary work he will
write a thesis that has practical use. I could also imagine that a thesis that produces a good metric for turning SRS raw data into a good metric would be cited. If we get a lot of people to use Anki and have a proper metric that could lead to published papers that will cite the metric.

In case you need data with has another cognitive test, I do use Anki and have a daily cognitive metric. I think that the same is true for Seth and Nick Winter.

1 Like

Can you explain why?

[quote]Can you explain why?[/quote]The median is a more robust metric. Bad measurements have less effects on the score.
Let’s say the phone rings in the background will you do Anki.
You take 20 seconds to grab it and put it on silent. Those 20 seconds won’t change the median much.
They will however have more effect on the average.

The same goes for the effects of single cards that encourage you to think much about them.

As a result the psychometrics people advocate the usage of medians over averages. I don’t have a background in that area but I think they have experimental test results to back themselves up.

Ah, it sounds like you’re more interested in speed of answering.

What I am tracking is % correct. What is most important to me is developing long-term memories of the material so that it becomes like second nature to me.

You’re right that sometimes cards will be more or less difficult. I actually try to break my cards up into the smallest pieces that I can so that they will be more or less equally difficult. For example, there’s a new drug called “Omacetaxine mepesuccinate” that I want to learn. I know that Omacetaxine mepesuccinate is going to be very difficult to memorize altogether, because it doesn’t resemble anything I know already & it has something like 10 syllables. Instead of putting the name in as one chunk, I test it as “… mepesuccinate” [answer: Omacetaxine]. Once I get that one correct enough times that my review time is pushed out to 3 months, I will chunk it together & start learning it as a bigger piece.

One concern I have with spaced repetition software is that they delink knowledge into chunks, making it harder to call them up when you need them.

For example, a person might know how to do some algebra problem but he does not know how to apply them to real life.

This isn’t really the problem SRS is designed to solve.

To be able to apply the algebra problem to real life, he must first know how to do it, right?

Yes, which is why I still do khanacademy(it uses SRS, apparently) and use anki for the rest.

I been trying to find a plausible solution to compartmentalized knowledge. So far concept mapping seems to be a promising tool. I saw one paper that describes high school students getting a higher score with concept mapping.

[quote] I actually try to break my cards up into the smallest pieces that I can so that they will be more or less equally difficult. [/quote]Of course. Part of being good at Anki is about being good at splitting up cards into small pieces.
Still sometimes you can’t and get harder cards.

[quote]Once I get that one correct enough times that my review time is pushed out to 3 months, I will chunk it together & start learning it as a bigger piece.[/quote]What’s your workflow for doing this?

I’m not sure I understand your request. I’m going to take a stab at it, but if I’m wrong, please let me know.

Each morning I have scheduled about an hour to do my daily reviews. I will sit down with no distractions & work through them all. The way Anki is setup, the most “mature” cards (those with the longest time-to-next-review) are presented first. If time-to-next-review is greater than or equal to 3 months (the green button on Anki for iPhone) AND I get the question right, I will go into Edit & copy the contents of the card. In most cards there are at least 2 things being “linked” together. For example, many of my cards are Brand & Generic names of drugs. In this case, I will often reverse the card – if it gave me the Brand & asked for the Generic, I will have a new card give me the Generic & ask me for the Brand.

Sometimes cards are more complicated, linking together many different things. In these cases, I will just choose what I consider to be the next most important element & convert that into the blank I need to fill in.

I might start with the card:

Q: In Ascites, daily Sodium intake should be limited to […] mEq or 2000mg.
A: 88

then once this card is “mature” past the 3 month time-to-next-review, I will make this card:

Q: In Ascites, daily Sodium intake should be limited to 88mEq or […] mg.
A: 2000

and once that one is “ripe”, finally add:

Q: In […], daily Sodium intake should be limited to 88mEq or 2000mg.
A: Ascites


I have heard about people doing it differently, using automated processes to make all of the cards all at once. I think my method, while more labor intensive, is superior. Each card reinforces the one that came before it. Each card is training access not only to the chunk it specifically tests for, but also to the other chunks that came before it.

Hope this is the question you were trying to ask & not rambling nonsense!

Cheers,

  • Adam
1 Like

[quote]I think my method, while more labor intensive, is superior. […] Hope this is the question you were trying to ask & not rambling nonsense!
[/quote]Yes, this is the question that I was asking. Many people use Anki differently so it’s interesting if people share their method of doing things.

If I understand you right you basically switch from answering cards to making a new card when the “ripe card” comes up.

I think there a huge cost when you have to make a decision for every card that you review: “Does this card I’m just answering is ripe enough that I can add another card to it, are there more cards that I have to add to that card?”

It doesn’t feel compatible with my approach taking 4 seconds per card review. Maybe *someday *Anki will support a way to create the card according to your rules with less manual labor.

I just finished a randomized experiment showing that walking on a treadmill did not enhance my Mnemosyne studying, but damaged it: http://www.gwern.net/Treadmill#treadmill-effect-on-spaced-repetition-performance-randomized-experiment

1 Like