* Fix/re-optimize FSRS if short-term param is weird
* Reset progress when another run is required (dae)
* only count the same-day steps
* Fix flicker when optimizing again (dae)
* Subtract introduced:1 count from learn count of first day of simulator.
* Fix: Cards filtered
* Tidy up
* Fix: Cards filtered (2)
* ./check
* Removed unnecessary filter
* Fix: Doesn't work for real new cards
* ./check
* Fix: .is_none()
* Limit to 1 day
* Removed "days_to_simulate" argument from convert
* Feat/support new cards ignore review limit in simulator
* ./ninja fix:minilints & ./ninja format
* use published crate
* make newCardsIgnoreReviewLimit reactive
* format
---------
Co-authored-by: Damien Elmes <gpg@ankiweb.net>
* new easy days algorithm
* take easy day percent totals in account when determining reduced scheduling
* Use variant method to avoid repeated mapping to a constant (dae)
It was probably not worth the time I took to change this ^_^;
* Add some unit tests
Covers most of the cases encountered in https://github.com/ankitects/anki/pull/3639
* Format
* Update params.rs
Makes the test more robust.
* Update params.rs
When training, the first FSRS item is removed. That's why none of the other tests includes it.
Co-authored-by: Jarrett Ye <jarrett.ye@outlook.com>
* Improve naming
* Fix typo
---------
Co-authored-by: Jarrett Ye <jarrett.ye@outlook.com>
* Fix/fallback to non-manual entry when first_of_last_learn_entries non found
* refactor single_card_revlog_to_item(s)
* update unit test of bypassed_learning_is_handled
* move comment line
* remove first_relearn_entries
* skip cram entry
* only pick non_manual_entries after ignore date
* fallback to non_manual_entries if the first learning step is before the ignore date
* pass ci
* update ignore_before_date_between_learning_steps_when_reviewing
* shorten the comment
* Minor refactoring
- fsrs_items_for_memory_state - fsrs_items_for_memory_states
- single_card_revlog_to_item -> fsrs_item_for_memory_state
(to match fsrs_items_for_training)
- single_card_revlog_to_items -> reviews_for_fsrs
- Use struct instead of tuple for reviews_for_fsrs output
- Don't return count, since we're already returning the filtered list
* More renaming/comment tweaks
- non_manual_entries -> first_user_grade_idx
- change comments to reflect the fact that we're working backwards
- Use "user-graded" rather than "non-manual"
* Add extra unit test
* Some wording tweaks
Because of the `entry.button_chosen == 0` part, it also filters out reschedule entries. Frankly, we don't need the `(entry.review_kind == RevlogReviewKind::Rescheduled)` part, but I didn't remove it yet.
Co-authored-by: Abdo <abdo@abdnh.net>
* Update to FSRS-rs v1.3.2
* add fsrs_short_term_with_steps_enabled to config
* ./ninja fix:minilints
* fix defaults_for_testing
* if current parameters are invalid, skip comparison
fix#3498
* fix redundant_field_names
* cargo clippy --fix
* Update to FSRS-rs v1.3.3
* Update to FSRS-rs v1.3.4
* Avoid an extra config lookup on each card answer (dae)
* Feat/export dataset for research
* add comment
Co-authored-by: Damien Elmes <dae@users.noreply.github.com>
* target_path is required
* format
* improve efficiency to look up parent_id
* move `use` down
* Feat/Easy Days
* format
* add easy_days_percentages to deck_config
* configure Easy Days via table
* remove unused code
* add translatable strings & add default of easy days
* don't check easy_days_percentages when deserialize
* pass test::all_reserved_fields_are_removed
* consider next_day_at when interval_to_weekday
* remove y-axis-title created in last simulation
* EstimatedTotalKnowledge should be integer
* Reorder deck option sections (dae)
- Move FSRS to bottom left, to move it closer to the top, and so
the left and right columns appear roughly balanced when FSRS is
enabled.
- Move Easy Days above Advanced
* Don't crash if wrong number of days (dae)
* Use lower field number (dae)
Repeated fields are more compactly stored in the first 15 fields.
* graduate card when user press hard and has 0 learning steps
* fix error: useless conversion to the same type
* do the same thing to again
* fix expected `Option<u32>`, found integer
* ./ninja format
* let FSRS control short term schedule
* Update to FSRS-rs v1.3.0
* ./ninja check:clippy
* Update to FSRS-rs v1.3.1
* Pin FSRS version (dae)
https://github.com/ankidroid/Anki-Android-Backend/pull/417
* Remove redundant parens (dae)
* keep the same-day reviews in feature
* Update to FSRS-rs v1.2.3
* format
* don't remove short-term reviews if not training
* Update to FSRS-rs v1.2.4
* Bump Python deps
Primarily for flask-cors CVE
* Bump TS deps; pin license checker
Current checker is missing the binary
https://github.com/RSeidelsohn/license-checker-rseidelsohn/issues/118
* Update Rust deps
Hyper and axum are held back as we currently make use of the older
http library that reqwest pulls in
* graduate card when user press hard and has 0 learning steps
* fix error: useless conversion to the same type
* do the same thing to again
* fix expected `Option<u32>`, found integer
* ./ninja format
* Update to FSRS-rs v1.2.0
* if else -> match
* Weight length check has been moved into FSRS (dae)
* Don't mention the number of FSRS parameters (dae)
It has changed, and may change again.
* test using existed cards
* plot new and review
* convert learning cards & use line chart
* allow draw multiple simulations in the same chart
* support hide simulation
* convert x axis to Date
* convert y from second to minute
* support clear last simulation
* remove unused import
* rename
* add hover/tooltip
* fallback to default parameters
* update default value and maximum of deckSize
* add "processing..."
* fix mistake
* start of load balancer
* add configuration options; option to load balance per deck
* formatting
* clippy
* add myself to contributors
* cleanup
* cargo fmt
* copyright header on load_balancer.rs
* remove extra space
* more formatting
* python formatting
* ignore this being None
only doing this cause python has awful lambdas and can't
loop in a meaningful way without doing this
* only calculate notes on each day if we are trying to avoid siblings
* don't fuzz intervals if the load balancer is enabled
* force generator to eval so this actually happens
* load balance instead of fuzzing, rather than in addition to
* use builtin fuzz_bounds rather than reinvent something new
* print some debug info on how its load balancing
* clippy
* more accurately load balance only when we want to fuzz
* incorrectly doublechecking the presence of the load balancer
* more printfs for debugging
* avoid siblings -> disperse siblings
* load balance learning graduating intervals
* load balancer: respect min/max intervals; graduating easy should be at least +1 good
* filter out after-days under minimum interval
* this is an inclusive check
* switch load balancer to caching instead of on the fly calculation
* handle case where load balancer would balance outside of its bounds
* disable lb when unselecting it in preferences
* call load_balancer in StateContext::with_review_fuzz instead of next to
* rebuild load balancer when card queue is rebuilt
* remove now-unused configuration options
* add note option to notetype to enable/disable sibling dispersion
* add options to exclude decks from load balancing
* theres a lint checking that the link actually exists so I guess I'll add the anchor back in later?
* how did I even update this
* move load balancer to cardqueue
* remove per-deck balancing options
* improve determining whether to disperse siblings when load balancing
* don't recalculate notes on days every time
* remove debug code
* remove all configuration; load balancer enabled by default; disperse siblings if bury_reviews is set
* didn't fully remove caring about decks from load balancer sql query
* load balancer should only count cards in the same preset
* fuzz interval if its outside of load balancer's range
* also check minimum when bailing out of load balancer
* cleanup; make tests happy
* experimental weight-based load balance fuzzing
* take into account interval when weighting as it seems to help
* if theres no cards the interval weight is just 1.0
* make load balancer disableable through debug console
* remove debug prints
* typo
* remove debugging print
* explain a bit how load balancer works
* properly balance per preset
* use inclusive range rather than +1
* -1 type cast
* move type hint somewhere less ugly; fix comment typo
* Reuse existing deck list from parent function (dae)
Minor optimisation
* Feat/FSRS-5
* adapt the SimulatorConfig of FSRS-5
* update parameters from FSRS-4.5
* udpate to FSRS-rs v1.1.0
* ./ninja fix:minilints
* pass ci
* update cargo-deny to 0.14.24
* udpate to FSRS-rs v1.1.1
* update to fsrs-rs v1.1.2
* Apply fuzz to SM2 lapse interval and respect max ivl
Imo, there is no reason for not applying fuzz to SM2 lapse intervals
* Update review.rs
* Format
* Update review.rs
* Update review.rs
* Update review.rs
* Update review.rs
* Fix FSRS easy interval being same as good interval in relearning cards
https://github.com/ankitects/anki/pull/3236#issuecomment-2187787774
* Update relearning.rs
* Update relearning.rs
* Set min interval of easy to Good + 1
* Ensure minimum doesn't exceed maximum (dae)
With a maximum interval set, it would be possible to confuse with_review_fuzz()
by passing min > max.
* return current weights if fsrs_items is zero
* handle error of evaluation if items.is_empty()
* TODO: handle error of optimal retention
* Revert "TODO: handle error of optimal retention"
This reverts commit 80a5b3803e.
* Revert "handle error of evaluation if items.is_empty()"
This reverts commit 7f0a5570e7.
* update fsrs to 0.4.5
* update to fsrs 0.4.6
* add benchmark API
* update fsrs to 0.5.0
* cargo fmt
* ./ninja fix:minilints
* ./ninja format
* Add a brief comment about the new method (dae)
* Fix/set pretrain_only and insufficient reviews based on fsrs_items
* use review_count for FsrsInsufficientReviews
* Update weights.rs
* Update weights.rs
- We must use interval, not stability to infer days_elapsed
- We must use original due date in a filtered deck
- Use retrievability in filtered deck sorting, not just regular sorting
Caused by using due instead of original_due when card was in learning.
I think the original goal of that code was to ignore the learning timestamp
and show the next review date instead, but it's both simpler and more
intuitive to show the learning date instead.
