SuperMemo Algorithm FAQ
(Redirected from Algorithm SM-17 FAQs)
- Question: How much will users benefit from Alg-SM17?. Answer: Is Algorithm SM-17 much better than Algorithm SM-15?
- When will I be able to use Alg-SM17?
- Will I have an option to use the old algorithm (Alg-SM15) in my collection? This will be possible in some early releases of the algorithm in SuperMemo for Window. However, the new algorithm is intended to replace all prior variants of the algorithms used in all SuperMemos and in all products. The old algorithm will be retained for a while for research purposes only
- Will the algorithm be open source, e.g. like Alg-SM2? No. The algorithm forms the basis of commercial activities of SuperMemo World and some of its core equations form a trade secret.
- Is it possible that Alg-SM17 would perform worse despite all the theory, e.g. due to a bug? Rigorous testing will ensure that the new algorithm performs better than older variants. Bugs are naturally inevitable and some may find their way to the final product. However, the algorithm will not see the light until it meets all metrics in a large group of users. The set of monitoring tools should be rich enough for you to inspect the process by yourself and make sure the algorithm meets all the design criteria. In SuperMemo 17 you will see a definite metric that will compare both algorithms live while learning
- Will the algorithm adapt to my memory or is it a universal formula? All SuperMemos adapt to your memory and, importantly, to your strategies (e.g. choice of grades). That adaptability will be stronger than ever. All that despite the fact that core formulas are now pretty universal due to being based on a strong memory model
- Will SuperMemo also change as a result of employing a new algorithm? No. SuperMemo does not need to change beyond a few technical details (like storing repetition histories). However, various versions are likely to be later equipped with new options that are made possible with the new algorithm (the array of possibilities in near-to infinite)
- If the S:R model is so simple, why do you speak of thousands lines of code? See: Why simple model leads to complex computer code?
- Can I switch my old collection to the new algorithm?
- Is stability just an interval? Why not to call it an interval?. See: Is stability just an interval?
- Will not new SuperMemo be fooled by outside world interference when using full repetition history record? No. All SuperMemos are constantly undermined by interference. Algorithm SM-17 provides the best statistical approach to look for regularities in the departure of data from memory models. See: Repetition history and outside interference
- Why do I have totally different intervals for the same item after a collection merge?
- Your formulas show that you can calculate R from S. Should this not mean that there is only one component of memory? No. See: How many components of memory are there?
- Once you say there are two components of memory. Once you say there are three. Can you explain? See: How many components of memory are there?
- If the two-component model is 22 years old, why do you come up with the algorithm only now?. See: Why does Algorithm SM-17 come so late?
- Should we not have two different measures of difficulty?. See: We should have two independent measures of difficulty!
- Did you try univalent matrices like in your 1994 publication?. See: How does Algorithm SM-17 perform when initialized with univalent matrices?
- Did you compare SM17 with older algorithms like SM2, Anki, SM8? Do you have specific numbers?. See: Algorithm SM-17 vs. older SuperMemos and SuperMemo 17 will use both Alg-SM16 and Alg-SM17
- Case against cramming - where strength of Algorithm SM-17 reveals weakness of human memory and fires back at the algorithm
- Why no impact of grades on the next interval?
- Might fixed retrievability be suboptimal?
- Why not base difficulty on startup stability
- How will you solve the problem of user-determined stability?
- How can stability increase almost linearly with the interval?
- Can you use R-metric to compare Algorithm SM-2 with Algorithm SM-17?
- What if my R-metric is negative?
- How can I interpret the stability increase matrix?
- What is the interpretation of New, Int17, and S in Algorithm SM-17?
- Shocking interval discrepancies
- Huge interval discrepancies between Algorithm SM-15 and SM-17
- Major interval discrepancies between SM16 and SM17 algorithms
- How can startup stability be two months? (incl. "ridiculous" 30-year intervals)
- SuperMemo 17 makes old SuperMemo look like "cramming tools"
- How often should I recompute the stability matrix?
- What is the impact of Forget and Memorize on stability?
- Why don't I get interval=1 in SuperMemo 17?
- Interpretation of stability column in Excel
- First interval is 2 months?
- Can I shorten repetition history in Algorithm SM-17?
- What is the difference between Algorithm SM-17 and the DSR model?
- Ignoring incomplete repetition histories
- How to interpret S, R, Next S, Next R, Int17, Int 18, etc?
- Error looking for optimum difficulty and startup stability
- Implementing my own algorithm with forgetting curves