Skip to main content

PublicHyperopt

The lean, anonymous, allowlisted projection of a FROZEN hyperopt result. This is the ENTIRE public surface of a shared hyperopt — it reveals the optimization STORY, never the strategy IP.

REVEALS (all safe): the loss-function name, the epoch count, the optimization SPACES, the tuned dimension NAMES (parameter names only), a normalized convergence curve (best-score-per-epoch, min-max scaled to a 0..100 SHAPE), and a relative score-improvement percent.

NEVER (structurally excluded): the won parameter VALUES (params_dict values — the crown-jewel IP), any absolute capital/stake/money, the raw result JSON / epochs_data / logs / config, owner_id, resource UUIDs, owner PII, or secrets. Built field-by-field; the ent.Hyperopt entity is NEVER embedded.

type PublicHyperopt {
  projectionLevel: ShareProjectionLevel!
  strategyName: String!
  lossFunction: String!
  epochs: Int!
  spaces: [String!]!
  tunedDimensions: [String!]!
  convergence: [Float!]!
  scoreImprovementPct: Float
  timeframe: String
  startDate: Time
  endDate: Time
}

Fields​

PublicHyperopt.projectionLevel ● ShareProjectionLevel! non-null enum​

The publisher's chosen projection level (card = teaser, full = all safe fields).

PublicHyperopt.strategyName ● String! non-null scalar​

Parent strategy human display name (never the mangled Freqtrade class name).

PublicHyperopt.lossFunction ● String! non-null scalar​

Human loss-function label (e.g. 'Sharpe Ratio') — never a value.

PublicHyperopt.epochs ● Int! non-null scalar​

Number of optimization epochs run.

PublicHyperopt.spaces ● [String!]! non-null scalar​

Optimization spaces (e.g. ['buy','sell','roi']) — high-level categories, never values.

PublicHyperopt.tunedDimensions ● [String!]! non-null scalar​

Tuned dimension NAMES (the optimized parameter names, e.g. ['buy_rsi','sell_rsi']), sorted. NAMES ONLY — the won VALUES are the strategy IP and are never included.

PublicHyperopt.convergence ● [Float!]! non-null scalar​

Best-score-per-epoch convergence curve, min-max normalized to a 0..100 SHAPE (0 = the first best-so-far, 100 = the final best). Reveals only how fast the search converged — never the absolute loss values. Empty at the card tier or when per-epoch data is unavailable.

PublicHyperopt.scoreImprovementPct ● Float scalar​

Relative improvement of the best-so-far loss from the first evaluated epoch to the final best, as a percent. A RELATIVE measure — never an absolute score or money. Null when it cannot be computed safely (sign change / zero baseline).

PublicHyperopt.timeframe ● String scalar​

Chart timeframe (e.g. '1h') from the frozen config.

PublicHyperopt.startDate ● Time scalar​

Optimization window start (honesty guardrail).

PublicHyperopt.endDate ● Time scalar​

Optimization window end (honesty guardrail).

Returned By​

publicHyperopt query