SpacedRepetition

Introduction

This package provides everything necessary to create spaced repetition software using a number of different popular algorithms. Information on the various algorithms available and the general structure of the package is provided below.

Algorithms

This package provides four different algorithms, three of which are based on the SM-2 scheduling algorithm. If you're not sure which to use, the algorithm provided by SMTwoAnki (which reproduces the scheduling behavior of the popular F/OSS flashcard software Anki) is most likely the "best" option.

Leitner (SpacedRepetition.Leitner)

The Leitner system was proposed by Sebastian Leitner in the early 1970s and was originally intended for use with physical (paper) flashcards. For the basics about this algorithm, please refer to the following description on Wikipedia.

In general, the Leitner system is much more simple than the other options provided (given that it was meant to be workable by hand) and should generally be considered deprecated in favor of the more advanced SM-2 based algorithms.

SM-2 (SpacedRepetition.SMTwo)

The SM-2 algorithm was one of the earliest computerized implementations of a spaced repetition algorithm (created in 1988 by Piotr Wozniak) and has been released for free public use when accompanied by the following notice:

Algorithm SM-2, (C) Copyright SuperMemo World, 1991.

• http://www.supermemo.com
• http://www.supermemo.eu

For details about this algorithm, please refer to the following description, written by its creator.

The SM-2 algorithm is robust, if more rudimentary than Anki's variant. Still, it may be useful for those desiring a simple system without settings that can still adapt to individual card difficulties (unlike the Leitner system).

SM2+ (SpaceRepetition.SMTwoPlus)

The SM2+ algorithm was proposed by "BlueRaja" as an improvement of the SM-2 algorithm. For details about the SM2+ algorithm and its purported advantages over the SM-2 algorithm, please refer to the following blog post.

It should be noted that this algorithm produces seemingly illogical behavior, namely that more incorrect answers result in longer intervals than less incorrect answers. In general, this algorithm has serious flaws as presented in its reference implementation and its superiority to the SM-2 algorithm is dubious at best. Nevertheless, it is implemented here as it is popular and often-cited online.

Anki (SpaceRepetition.SMTwoAnki)

The algorithm used by the popular F/OSS program Anki, this algorithm is a heavily-modified version of the SM-2 algorithm. For details about Anki's algorithm, please refer to the following section of its manual.

This is by far the most powerful and flexible algorithm provided in this package and should be considered the "default" for most users.

General Use

The following functions/types are provided by every algorithm:

Card and Deck

The building blocks of this package are Cards and Decks. In simple terms, a Card may be thought of as a single flashcard and a Deck as a list or collection of Cards. Card is always defined in terms of an extensible record and contains only the data necessary for scheduling, so that the user of this package may add whatever fields they find necessary for actually holding data on the card.

SRSData and newSRSData

SRSData contains all information necessary for scheduling a card. In all cases, a Card may be created by use of the newSRSData function, as in the following example:

type alias MyFlashcard =
    Card { prompt : String, answer : String }

myFlashcard : MyFlashcard
myFlashcard =
    { prompt = "SYN"
    , answer = "SYN-ACK"
    , srsData = newSRSData
    }

encoderSRSData and decoderSRSData

All algorithms provide Json encoders and decoders for SRS data, which may be utilized as follows:

import Json.Decode as Decode
import Json.Encode as Encode

type alias MyFlashcard =
    Card { prompt : String, answer : String }

myFlashcardConstructor : SRSData -> String -> String -> MyFlashcard
myFlashcardConstructor srsData prompt answer =
    { prompt = prompt
    , answer = answer
    , srsData = srsData
    }

myFlashcardToJson : MyFlashcard -> String
myFlashcardToJson myCard =
    Encode.encode 0 <|
        Encode.object
            [ ( "srsData", encoderSRSData myCard.srsData )
            , ( "prompt", Encode.string myCard.prompt )
            , ( "answer", Encode.string myCard.answer )
            ]

myFlashcardDecoder : Decode.Decoder MyFlashcard
myFlashcardDecoder =
    Decode.map3 myFlashcardConstructor
        (Decode.field "srsData" decoderSRSData)
        (Decode.field "prompt" Decode.string)
        (Decode.field "answer" Decode.string)

jsonToMyFlashcard : String -> Result Decode.Error MyFlashcard
jsonToMyFlashcard str =
    Decode.decodeString myFlashcardDecoder str

answerCardInDeck and answerCard

When a card is presented to the user and answered, answerCardInDeck should be called. It always takes the current time (in the Time.Posix format returned by the now task of the core Time module), some sort of answer or performance (the Answer type for all algorithms except SM2+), the index of the card in the Deck, and the Deck itself. It returns the updated Deck. Use this function if you simply want to store a Deck and not worry about updating it manually (which is most likely what you want). Otherwise, use answerCard to handle updating the Deck manually.

getDueCardIndices

getDueCardIndices takes the current time (in the Time.Posix format returned by the now task of the core Time module) and a Deck and returns the indices of the subset of the Deck that is due for review. The sorting of the results varies with the algorithm.

QueueDetails, getCardDetails, getDueCardIndicesWithDetails

If you need information about the SRS status of a card (e.g. when it was last reviewed, whether it's new, etc.), such information may be found in the QueueDetails of a module. QueueDetails may be obtained from a single Card with getCardDetails or along with the indices of due cards with getDueCardIndicesWithDetails.

Miscellaneous

The various algorithms provide additional functions/types as necessary for their individual implementations. Refer to their documentation for specifics.

Changelog

• 2.1.0
  • Actual JSON encoding has not changed, so this is compatible with JSON generated by 2.0.1, but validation is more strict on things that should be non-negative (which would not normally have been written by 2.0.1, so it should not cause issues).
  • 🐛 Bugfix: For all algorithms, equivalently-due cards would appear in reverse input order. This shuffles their order instead, to prevent the same ordering from occurring repeatedly.
  • SpacedRepetition.Leitner
    • 🏷️ NumberOfBoxes is now exposed (but still opaque), so you may write type signatures with it.
    • 📝 Note that cards will be graduated after answering (even with Pass) if they're in an invalid box beyond NumberOfBoxes. This was always the case, but it's mentioned in the documentation now.
    • 🐛 Bugfix: Enforce SpacingFunctions returning an interval of at least 1 day; this was always the case per documentation and you definitely couldn't cause problems by returning zero prior to this version (shh...).
    • ⚡️ Tail-call optimized fibonacciSpacing, so you can have intervals of 10^38 years for your 200th box if you're an eternal but not omniscient being.
  • SpacedRepetition.SMTwoAnki
    • 🐛 Bugfix: Ensure that cards always graduate from being "lapsed" regardless of the answer if there are no lapse steps. This was the behavior specified in the documentation of lapseSteps, but it wasn't actually happening. Now, answering Hard or Again on a "lapsed" card will return it to the review queue if there are no lapse steps.
      • Old behavior: With no lapse steps, failing a review card will lapse it, making it immediately due for review. Answering Hard or Again will leave it immediately due. Answering Good or Easy will return it to the review queue.
      • New behavior: With no lapse steps, failing a review card will lapse it, making it immediately due for review. Answering the card with any answer will return it to the review queue.
    • 🐛 Bugfix: Ensure that cards always graduate from learning regardless of the answer if there are no learning steps. This was the behavior specified in the documentation of newSteps, but it wasn't actually happening. Now, answering Hard or Again on a learning card will graduate it to the review queue if there are no learning steps. Unlike the case with lapses, however, this bug should have been quite rare in practice, as the only way to end up with cards in the learning queue with no learning steps would be to change the settings of a deck that had already been partially studied to remove previously-extant newSteps.
    • 🚸 getLeeches now returns cards with the same number of lapses in order of their appearance in the input deck.
    • 🏷️ TimeInterval, Days, and Minutes are now exposed (but still opaque), so you may write type signatures with them.
    • 📝 Fix broken links to Anki documentation, since the URLs moved.
    • 🩹 Fix bug in non-exposed function. This bug could not have actually caused erroneous behavior in any exposed functions, but it might have going forwards had its output been used for something else.
• 2.0.1 -- 🐛 Fixed a bug in SpacedRepetition.SMTwoAnki that caused the extra interval from studying an overdue card to not count with Good answers. Per the algorithm, half of the overdue amount should be included in calculating the new interval with a Good answer.
• 2.0.0 -- Added getDueCardIndicesWithDetails and getCardDetails to all modules, allowing one to get information about e.g. what stage of learning a card is in so that it might be displayed differently. This was a breaking change because the return type of SpacedRepetition.SMTwoAnki.getDueCardIndices changed to no longer return leech status (use getDueCardIndicesWithDetails or getCardDetails to get leech status).
• 1.1.0 -- Added a JSON encoder/decoder for SpacedRepetition.SMTwoAnki.AnkiSettings
• 1.0.0 -- Initial release.