fasiha / ebisu-java

A Java port of Ebisu, the Bayesian quiz scheduling library.

• 👍 he/him (they/them 👌)
• 🔭 do check out these newer projects
  • Yoyogi is an anti-timeline frontend for Mastodon that centers a single author and their threads. I'd been asking others to make this for years and finally did it 😝
  • Kamiya Codec conjugates and deconjugates Japanese verbs and adjectives following Taeko Kamiya's texts
  • Hareonna visualizes temperature percentiles from 13'500 stations in the Global Historical Climatology Network daily (GHCNd) database. Find where your perfect weather is!
  • Devastate is a web app version of the game to calibrate uncertainty from Julia Galef's The Scout Mindset
  • HTML Ruby Tag is a VS Code extension for ルビruby-tag-lovers.
  • Yamanote, a bookmark annotation and archival system with browser-based snapshots, meaning it works great with paywalls, JavaScript-heavy sites, etc.!
  • Midsummer's Night Dream script of the 2010 Original Pronunciation (OP) abridged radio play by Kansas Public Radio and University Theatre at the University of Kansas. For audio and script PDF, see Paul Meier
  • CJK Text Processing and Languages, my talk for an informal lunch-and-learn where I fanboy Chinese, Japanese, and Korean
  • Gotanda, a centralized backup sync target for all your PouchDB-powered local-first apps
  • Kanda for intensive Japanese reading and review leveraging powerful Japanese NLP
  • Curtiz combines super-powered Japanese NLP tools in Node: MeCab, UniDic, J.DepP, JMDict, JMDict-Furigana
  • Texture-shaded Globe webapp and implementation of the lovely terrain-shading algorithm
  • Call Faye is the perfect way to play probabilistic Texas hold 'em poker with the kids. See the blog post.
  • MemriseDriver: add text-to-speech to your Memrise flashcards, via AWS Polly (Japanese support!) and Puppeteer
• 🤳 long-time favorites
  • Ebisu for scheduling flashcards with Bayesian statistics
  • Mudder.js: subdivide lexicographic space between strings (also base62)
  • KanjiBreak: crowdsourced breakdowns for Chinese characters
  • KanjiVG Explorer: explore Chinese character dependency graphs
• 🔍 Not big enough to be projects—more like amuses-bouches:
  • Giant satellite-projected maps for your walls, a set of four or seven: print them two meters wide!
  • Frequency table of leading sounds in syllables in Taigi (Taiwanese Hokkien) vs Cantonese vs Mandarin
  • Hear and see the same sentence in all sixty-five languages currently supported by Glossika
  • Word search puzzle for you homeschoolers—Japanese or English, excellent printer support!
  • US Covid deaths by county 2020 election results—morbid charts
• 🤙 contact—happy to chat, mentor, collaborate
• 🧑‍🎨 profile photo: Sun Wukong the Monkey King by the fabulous @monarobot in her Maya style
• 🇮🇳 always try the paan
• 🇨🇳 always try the mapo tofu
• 👄 very gentle pronunciation suggestion: "AAH-med FAH-see" (احمد فصيح for the Arabic/Persian readers)

ebisu-java's Issues

Update

Is this lib up-to-date compare to js and python counterparts? How do we keep track of the last change in python?

updateRecall api usage

My implementation is when a student learns a fact, I construct an Ebisu model. After t amount of time, I would call

var predictedRecall = ebisu.predictRecall(model, elapsed, true); elapsed here is computed by currenttime - ebisu created timestamp. This seems very straightforward. When it drops below threshold, I convert a fact into a quiz.

With updateRecall api

it requires an elapsed time. Based on your docs it is ebisu.updateRecall(model, result, tnow) -> model to update the model given a quiz result and time after its last review. what do you mean last review time? is it the same as createTimestamp when I constructor Ebisu model? Given it's a different timestamp, I will call it reviewTimestamp, when user reviews a fact, I create reviewTimestamp, after a minute he takes a quiz, elapsed would be = 0.1 in this case, my question is what is the value for elapsed if user doesn't review since he learned a fact?

How difficult would it be to port Ebisu to C#?

I have a platform that uses C# so it's a problem for me that there's no C# version. Do you have any ideas how difficult it would be to port the code to C#? Are there any obstacles such as the availability of libraries that you can think of?

Check if a fact is internalised

hey thank you for the library again. my question is not really related to Ebisu. After a student reviews a fact, does some quizzes to improve recall. How do we know if he internalises that fact he learn? Is there any algorithm out there, any way, any formula to achieve this?

C# port of Ebisu

Hi,

I have ported the Ebisu in C# from your repo which is in Java.
Here is the link to my repo :- https://github.com/nikhildpardasani/Ebisu-CSharp
Kindly go through the code, I have maintained good code. Do let me know if you feel any changes required in any part of the code.
This repo contains all three maximise golden function, gamma function and Ebisu.
Can you please verify this code if it works fine and if you approve that it is correct can you make a note in the Ebisu Readme with a link to the repo.

Thanks!

Predict negative number

On your readme, the second line return -0.69.... What do we do with negative number?

timeElapsed ==> 0.25

jshell> double logRecallProbability = Ebisu.predictRecall(model1, timeElapsed);
logRecallProbability ==> -0.6931471805599458

jshell> double recallProbability = Ebisu.predictRecall(model1, timeElapsed, true);
recallProbability ==> 0.4999999999999997

Requesting an update of the Java code to use the Ebisu 2.0 algorithm

I would like to start soon porting the code to C#. Once completed I would be happy to make this available to you so we could have another open source C# port.

When you have time available it would be a great help if you could port the code to the Ebisu 2.0 algorithm.

Thanks

Ebisu 1.0 new method usage

Based on the discussion here fasiha/ebisu#11 . Api method
predictRecall and updateRecallModel are mostly used. Since you updated to 1.0, there are some new methods like modelToPercentileDecay, how do we really use this inside a quiz app?