Math Practice

Math practice for kids (and anyone else) - counting, times table recitation, and interactive arithmetic quizzes across all four basic operations plus percentages, difficulty levels, chained and mixed-operator problems, an estimation mode, and one-decimal-place arithmetic. Fully offline.

• Four modes
• Difficulty
• Chained and mixed-operator problems
• Estimation mode
• Decimal arithmetic
• Usage
• Teach-then-practice
• How the quiz works
• Problem generation defaults
• Install
• Development

Four modes, deliberately not one "do math" intent

1. Counting - "count to ten" - pure recitation, no interaction.
2. Table recitation - "say the 5 times table" - pure recitation of 1×5 through 10×5, no interaction. Multiplication only - addition/subtraction/division don't have a traditional "table" concept the same way, so this stays scoped to what "times table" actually means.
3. Teach, then practice - "teach me the 5 times table" (or "teach me addition facts for 5" / "teach me subtraction facts for 5" / "teach me division facts for 5" - all four operations, not just multiplication) goes through the facts one row at a time (say "repeat" to hear a row again before moving on), then "quiz me on what you taught me" quizzes specifically on THOSE facts - not arbitrary random problems, and in the SAME operation that was taught. See "Teach-then-practice" below.
4. Quiz - "quiz me on the 3 times table" / "quiz me on addition" / "quiz me on percentages" - genuinely interactive: asks 5 questions one at a time, listens for a spoken answer, and reports a final score. Covers all four basic operations plus percentages, not just multiplication - this was an explicit scope correction from the original "times tables" framing during design.

"give me a math quiz" randomizes across the classic four operations only, same as it always has. "give me a full math quiz" randomizes across every operation this skill knows (currently the classic four plus percentages) - a separate, broader pool so new operations don't retroactively change what the classic phrase means. See the ALL_OPERATIONS / OPERATIONS module comment in __init__.py and issue #7 for the reasoning.

Difficulty

"quiz me on hard addition" / "quiz mig i svær division" selects one of easy / medium / hard for that quiz round only - there's no persisted skill-wide default (see issue #2). "quiz me on addition" with no difficulty word still works exactly as before - it runs at medium, which reproduces the original ranges every operation always used. See DIFFICULTY_RANGES in __init__.py for the actual per-operation, per-tier ranges. Times-table quizzing ("quiz me on the 5 times table") doesn't take a difficulty - the table number itself already controls how hard it is.

Chained and mixed-operator problems

"quiz me on chained addition" / "quiz mig i kædet subtraktion" asks a 3-operand, same-operator question ("what is 7 minus 2 minus 1") instead of the usual two-operand one. Subtraction and division chains are built so every INTERMEDIATE step stays valid (non-negative / evenly divides), not just the final result - see issue #3.

"quiz me on mixed operators" / "quiz mig i blandede regnearter" asks a 3-operand question with one +/- operator and one x/÷ operator, testing real order-of-operations precedence ("what is 4 plus 3 times 2" = 10, not 14) - see issue #4, including its open question about whether a spoken question conveys precedence as unambiguously as a written one does; worth verifying on a live instance rather than assuming.

Neither mode is difficulty-aware or part of OPERATIONS/ ALL_OPERATIONS for v1 - a deliberate, scoped-down choice, not an oversight.

Estimation mode

"quiz me on estimation" / "quiz mig i estimering" asks a large-number multiplication or division question with three lettered choices (A/B/C) instead of a computed spoken number - avoids requiring an unwieldy 8-digit answer to be spoken, and lets the user answer with either the letter or the number itself. The two wrong choices are built from specific, plausible estimation mistakes (a decimal-place slip; recomputing with one factor rounded to a nearby "nice" number), not random noise - see issue #8 and generate_estimate_problem() in __init__.py. Whether three choices is the right number for a voice-only interaction (vs. glancing at a screen) is flagged as an open question in the issue itself - worth testing on a live instance.

Decimal arithmetic

"quiz me on decimal addition" / "quiz mig i decimaltal i subtraktion" asks a one-decimal-place question ("what is 7.3 plus 2.5"), reusing the exact same quiz_question_<op>.dialog files as the integer version. Every decimal problem is constructed to be EXACT (integer-tenths arithmetic internally, converted to a float only once) rather than needing a genuine estimation-tolerance band - see issue #5 and generate_decimal_problem() in __init__.py. This sidesteps the tolerance-band question for THIS skill (every operation here can be constructed exact); it does not answer it for genuinely irrational quantities like ovos-skill-unit-practice's meter-to-mile conversion, which will still need real tolerance-band grading. Fractions are intentionally out of scope here - a separate, later design pass, per the issue's own recommendation. Whether STT reliably transcribes a spoken decimal number at all (e.g. "seven point three") is untested - flagged in the issue, worth verifying on a live instance.

Usage

"count to ten"
"say the 5 times table"
"teach me the 5 times table"
"teach me addition facts for 5"
"teach me subtraction facts for 5"
"teach me division facts for 5"
"quiz me on what you taught me"
"quiz me on the 3 times table"
"quiz me on addition"
"quiz me on percentages"
"quiz me on hard addition"
"quiz me on chained subtraction"
"quiz me on mixed operators"
"quiz me on estimation"
"quiz me on decimal addition"
"give me a math quiz"
"give me a full math quiz"
"tæl til ti"                    (Danish)
"sig 5 tabellen"                (Danish)
"lær mig 5 tabellen"            (Danish)
"lær mig plus for 5"            (Danish)
"quiz mig i det du lærte mig"   (Danish)
"quiz mig i 3 tabellen"         (Danish)
"quiz mig i procent"            (Danish)
"quiz mig i svær division"      (Danish)
"quiz mig i kædet subtraktion"  (Danish)
"quiz mig i blandede regnearter" (Danish)
"quiz mig i estimering"         (Danish)
"quiz mig i decimaltal i addition" (Danish)
"quiz mig i det hele"           (Danish)

Teach-then-practice

The reference implementation of the pattern shared across the whole *-practice family (see issue #1 for the original design discussion, since resolved and extended to all four operations). "teach me the 5 times table" speaks each row in turn, waits for either "repeat" (says the same row again) or anything else (moves to the next row), and records exactly which facts were presented AND which operation they belong to. "quiz me on what you taught me" then quizzes ONLY on that recorded set, in the same operation, reusing the same grading logic as the regular quiz mode (_ask_and_grade()), rather than generating fresh random problems.

Started multiply-only (mirroring "times table" recitation), then generalized to all four operations via a shared FACT_TABLE_GENERATORS registry - addition/subtraction/division each get their own "N-facts table" generator, mirroring multiplication's table shape (subtraction and division specifically constructed so results stay non-negative / evenly divide, matching the same rules the regular quiz mode already enforces).

Deliberately session-only for this release: self._taught_facts resets when the skill restarts - it doesn't persist across days. This was a scoped-down choice from the original design discussion, which also considered persisting taught facts (via self.settings or a local file) so "quiz me on what you taught me" would still work in a later session. Kept simple for v1; worth revisiting if session-only turns out to be a real limitation in practice rather than a theoretical one.

How the quiz works

Unlike ovos-skill-metronome/ ovos-skill-rhythm-box/ ovos-skill-white-noise, quiz mode needs no background thread - it's a sequential, blocking conversation using OVOSSkill.get_response(), which already handles the listen-and-transcribe round-trip. Each of the 5 questions is asked, answered (or timed out), and scored in turn; a timed-out or unparseable answer counts as incorrect rather than crashing the quiz.

Problem generation defaults

• 5 questions per quiz round.
• Times tables: the traditional 1-12 range.
• Addition/subtraction: operands 1-20, subtraction always kept non-negative (larger number first).
• Division: always divides evenly (built as divisor × quotient, both 1-10) - no fractional answers, since spoken fraction-checking is a much harder problem than this skill attempts to solve.
• Percentage: built backwards like division - the percentage (1-100) is picked first, then a base is constructed that's guaranteed to make percent/100*base a whole number, rather than picking both numbers freely and rounding. No tolerance/rounding needed - same "construct it exact" approach decimal arithmetic uses too (see "Decimal arithmetic" above); fractions are the one remaining piece of issue #5 that may genuinely need a tolerance band.
• All of the above are the medium difficulty ranges - see "Difficulty" above for easy/hard.

Install

pip install ovos-skill-math-practice

Development

See DEVELOPMENT.md.

Category

Education

Tags

#math #education #times-tables #quiz #kids