Product documentation — installation, licensing, and integration guides.
SolvScout / SolvTune
Examples
03 Recommend a solver

Journey 03 — Recommend a solver

Rank solver arms for a fixed ProblemProfile under different objectives (balanced, precision, survival, efficiency) and read dead zones.

Prerequisites: Journey 02 — you understand ProblemProfile. Scenario: One profile, several objectives, ranked arms

Key APIs

  • solvscout.recommend(profile, objective=...)
  • Recommendation.best
  • Recommendation.rankings
  • ArmRecommendation.dead_zone

Concepts to internalize

  • Objectives — balanced (default), precision, survival, efficiency. The top arm can change with your goal.
  • dead_zone — arms SolvTune will not recommend on this profile class, with a reason string.
  • SolverConfig on best — tuned knobs ready to map into solvsrk.run(); journey 05 shows the opaque token path instead.
  • Confidence — how close your problem is to benchmarked corpus entries.

Important code

The guided tour script wraps this pattern. Copy the production shape into your pipeline:

import solvscout as scout
 
rhs = ...  # Rössler from journey 02
profile = scout.classify(rhs, [1, 1, 1], (0, 100)).profile
 
for objective in ('balanced', 'precision', 'survival', 'efficiency'):
    rec = scout.recommend(profile, objective=objective)
    best = rec.best
    print(objective, best.arm if best else None, best.confidence if best else None)
 
# Full ranking with scores and dead-zone reasons:
rec = scout.recommend(profile, objective='balanced')
for r in rec.rankings:
    print(r.arm, r.score, r.confidence, r.dead_zone, r.dead_zone_reason)

Act-by-act walkthrough

Act 1 — Classify once, recommend many ways

recommend(profile, objective=...) ranks every active solver arm for the chosen goal. Objectives: 'balanced' (default), 'precision', 'survival', 'efficiency'.

FieldValueNotes
problemrossler
stiffnessS1

Act 2 — Compare objectives

The top pick can change with the objective - a survival-first choice is not always the most precise one. best is the top viable arm.

recommend() per objective (4 steps)

FieldValueNotes
best_armsolvsrk
confidencehigh
n_viable14
n_dead_zones0
best_armsolvsrk
confidencehigh
n_viable14
n_dead_zones0
best_armsolvsrk
confidencehigh
n_viable14
n_dead_zones0
best_armext_cvode_bdf
confidencelow
n_viable14
n_dead_zones0

Act 3 — Read the full ranking

Each ArmRecommendation carries a score, confidence, survival_rate, and a ready-to-use SolverConfig. dead_zone arms are flagged with a reason.

balanced ranking (14 steps)

FieldValueNotes
score0.7199
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.7186
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.6964
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.6903
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.6322
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.6216
confidencehigh
survival_rate1.0
dead_zoneFalse
reason-
score0.6031
confidencelow
survival_rate0.8
dead_zoneFalse
reason-
score0.5818
confidencelow
survival_rate0.75
dead_zoneFalse
reason-
score0.5691
confidencelow
survival_rate0.72
dead_zoneFalse
reason-
score0.5691
confidencelow
survival_rate0.72
dead_zoneFalse
reason-
score0.4968
confidencelow
survival_rate0.55
dead_zoneFalse
reason-
score0.4968
confidencelow
survival_rate0.55
dead_zoneFalse
reason-
score0.4968
confidencelow
survival_rate0.55
dead_zoneFalse
reason-
score0.494
confidencehigh
survival_rate0.75
dead_zoneFalse
reason-

Guided tour output

Verified console capture from 03_recommend.py:

One profile, several objectives, ranked arms

-- Act 1 - Classify once, recommend many ways --
  -> recommend(profile, objective=...) ranks every active solver arm for the chosen goal. Objectives: 'balanced' (default), 'precision', 'survival', 'efficiency'.
  problem: rossler
  stiffness: S1

-- Act 2 - Compare objectives --
  -> The top pick can change with the objective - a survival-first choice is not always the most precise one. best is the top viable arm.
  _section: recommend() per objective (4 steps)
  best_arm: solvsrk
  confidence: high
  n_viable: 14
  n_dead_zones: 0
  best_arm: solvsrk
  confidence: high
  n_viable: 14
  n_dead_zones: 0
  best_arm: solvsrk
  confidence: high
  n_viable: 14
  n_dead_zones: 0
  best_arm: ext_cvode_bdf
  confidence: low
  n_viable: 14
  n_dead_zones: 0

-- Act 3 - Read the full ranking --
  -> Each ArmRecommendation carries a score, confidence, survival_rate, and a ready-to-use SolverConfig. dead_zone arms are flagged with a reason.
  _section: balanced ranking (14 steps)
  score: 0.7199
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.7186
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.6964
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.6903
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.6322
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.6216
  confidence: high
  survival_rate: 1.0
  dead_zone: False
  reason: -
  score: 0.6031
  confidence: low
  survival_rate: 0.8
  dead_zone: False
  reason: -
  score: 0.5818
  confidence: low
  survival_rate: 0.75
  dead_zone: False
  reason: -
  score: 0.5691
  confidence: low
  survival_rate: 0.72
  dead_zone: False
  reason: -
  score: 0.5691
  confidence: low
  survival_rate: 0.72
  dead_zone: False
  reason: -
  score: 0.4968
  confidence: low
  survival_rate: 0.55
  dead_zone: False
  reason: -
  score: 0.4968
  confidence: low
  survival_rate: 0.55
  dead_zone: False
  reason: -
  score: 0.4968
  confidence: low
  survival_rate: 0.55
  dead_zone: False
  reason: -
  score: 0.494
  confidence: high
  survival_rate: 0.75
  dead_zone: False
  reason: -

-- Run complete --
  next_journey: 04_compare_arms.py
  best_arm: solvsrk

Run it

cd python
# Guided tour — narrated scenario walkthrough (recommended first run):
python examples/journeys/03_recommend.py
 
# Production one-liner — same API call you ship:
python -c "import solvscout as s; rhs=lambda t,y:(lambda x,z,w:[-z-w,x+0.2*z,0.2+w*(x-5.7)])(*y); c=s.classify(rhs,[1,1,1],(0,100)); print(s.recommend(c.profile).best.arm)"

What comes next

Journey 04 uses compare() for a full corpus-backed work-precision table.

Navigation

← Journey 02 — Classify an RHS · Journey 04 — Compare arms →

02 Classify an RHS04 Compare arms