[Question] How can I set the parameters in a way which can make the prediction closer to Anki's

[arungeorgesaji]

Right now this is what I have,so basically it sets the new values to the same as ankis and I set the same default weights aswell and if I continuosly press good 3 times it seems like maxes the time out there and then it just procceeds to change around only because fuzziness is turned on in anki it keeps increasing. and values changes for each rating keeps changing...How can I adjust the parameters to have a closer to result to that of anki am I doing something wrong??

import { fsrs, generatorParameters, Rating } from 'ts-fsrs';

export function initializeFSRS() {
    const params = generatorParameters({ 
        enable_fuzz: true, 
        request_retention: 0.9,
        maximum_interval: 36500,
        w: [0.4026, 1.1839, 3.1730, 15.6910, 7.1949, 0.5345, 1.4604, 0.0046, 1.5458, 0.1192, 1.0193, 1.9395, 0.1100, 0.2961, 2.2698, 0.2315, 2.9898, 0.5166, 0.6621],
        enable_short_term: true, 
    });
    return fsrs(params);
}


export function updateCard(scheduling, card, rating) {
    if (scheduling) {
        const { card: updatedCard } = scheduling;
        if (card.state == 0) {
            const now = new Date();

            switch (rating) {
                case 1:
                    updatedCard.due = new Date(now.getTime() + 1 * 60 * 1000); 
                    break;
                case 2:
                    updatedCard.due = new Date(now.getTime() + 6 * 60 * 1000); 
                    break;
                case 3:
                    updatedCard.due = new Date(now.getTime() + 10 * 60 * 1000); 
                    break;
                case 4:
                    updatedCard.due = new Date(now.getTime() + 13 * 24 * 60 * 60 * 1000); 
                    break;
                default:
                    console.error("Invalid rating:", rating);
                    return;
            }
        }
        Object.assign(card, updatedCard); 
    }
}

export function processCard(f, card, rating) {
    const scheduling = f.next(card, new Date(), rating);
    updateCard(scheduling, card, rating);
}

export function getCardInterval(f, card, rating) {
    const scheduling = f.next(card, new Date(), rating);

    const intervals = {
        1: "1 minutes",      
        2: "6 minutes",  
        3: "10 minutes",     
        4: "13 days",
    };

    if (card.state == 0) {
        return intervals[rating] || "Unknown";
    }
    
    if (!scheduling) {
        console.error("No scheduling found for rating:", rating);
        return "Unknown";
    }

    const newDueDate = new Date(scheduling.card.due);
    const now = new Date();
    const intervalSeconds = (newDueDate - now) / 1000; 

    if (intervalSeconds < 3600) {
        return `${Math.round(intervalSeconds / 60)} minutes`;
    } else if (intervalSeconds < 86400) {
        return `${Math.round(intervalSeconds / 3600)} hours`;
    } else {
        return `${Math.round(intervalSeconds / 86400)} days`;
    }
}

[ishiko732]

There are three differences:

• The learning steps in Ts-fsrs are currently hardcoded (state.New, state.(Re)Learning), whereas Anki has implementations for (re)learning, which ts-fsrs has not yet implemented( PRs are welcome ).
  • https://docs.ankiweb.net/deck-options.html#learning-steps
  • https://docs.ankiweb.net/deck-options.html#relearning-steps
• When enable_short_term is enabled, Ts-fsrs uses the state to schedule the short-term algorithm, while Anki calls the short-term algorithm when ivl=0. This can be resolved in ts-fsrs by using fsrs.nextState.
  • use fsrs.next_state to estimate memory state accurately Luc-Mcgrady/Anki-Search-Stats-Extended#23
• Starting from version 24.11, Anki has the “easy days” feature. https://github.com/ankitects/anki/releases/tag/24.11

[arungeorgesaji]

Thanks for the info.

[Math-Vieira]

Hey man, can you help me?
How can I change the first value of each review option?
I mean, they start like this:
again: 1 min
hard: 5 min
good: 10 min
easy: 2 weeks

How can I change the last one to a value that I assume is more correct?

[ishiko732]

Hey man, can you help me? How can I change the first value of each review option? I mean, they start like this: again: 1 min hard: 5 min good: 10 min easy: 2 weeks

How can I change the last one to a value that I assume is more correct?

There are currently no configurable parameters, but you can handle it using afterHandler.

Note: easy should not be less than 1 day.

example: (online: https://jsfiddle.net/ishiko/9vmgdys8/2)

import { createEmptyCard, fsrs, Grade, Rating, RecordLogItem, State } from 'ts-fsrs'

const learningSteps: Map<Grade, number> = new Map([
  [Rating.Again, 10 /** min */],
  [Rating.Hard, 30 /** min */],
  [Rating.Good, 60 /** min */],
])

const relearningSteps: Map<Grade, number> = new Map([
  [Rating.Again, 30 /** min */],
  [Rating.Hard, 60 /** min */],
])

function changeStep(item: RecordLogItem) {
  const grade = item.log.rating as Grade
  if (item.card.state === State.Learning) {
    const review: Date = new Date(item.card.last_review!)
    review.setTime(review.getTime() + learningSteps.get(grade)! * 60 * 1000)
    item.card.due = review
  } else if (item.card.state === State.Relearning && relearningSteps.has(grade)) {
    const review: Date = new Date(item.card.last_review!)
    review.setTime(review.getTime() + relearningSteps.get(grade)! * 60 * 1000)
    item.card.due = review
  }

  return item
}

const f = fsrs()
const now = new Date('2025-03-10T00:00:00.000Z')
const card = createEmptyCard(now)
const grade = Rating.Easy
const items_change_step = f.next(card, now, grade, changeStep)