HKUDS/Vibe-Trading · error · ValuationError

{model}: {name} must be a finite number, got {val!r}

Error message

{model}: {name} must be a finite number, got {val!r}

What it means

calendarise_metric validates that every fiscal-period value it uses to calendarise a metric (last_full_fiscal_year, current_year_to_date, prior_year_to_date, next_full_fiscal_year) is a finite number before combining them under an LTM/NTM policy. NaN or infinity in any period would silently poison the blended figure, so the library refuses it.

Source

Thrown at agent/src/quantlib/valuation/comps.py:273

        ValuationError: If `policy` is not one of `CALENDARISATION_POLICIES`,
            or if any supplied period value is not a finite number.
        MissingInputError: If `periods` lacks a field this policy needs.
    """
    if policy not in CALENDARISATION_POLICIES:
        raise ValuationError(
            f"comps.calendarise_metric: unknown policy {policy!r}, must be "
            f"one of {CALENDARISATION_POLICIES}"
        )
    model = f"comps.calendarise_metric[{company_name}.{metric_name}:{policy}]"

    for name, val in (
        ("last_full_fiscal_year", periods.last_full_fiscal_year),
        ("current_year_to_date", periods.current_year_to_date),
        ("prior_year_to_date", periods.prior_year_to_date),
        ("next_full_fiscal_year", periods.next_full_fiscal_year),
    ):
        if val is not None and not math.isfinite(val):
            raise ValuationError(
                f"{model}: {name} must be a finite number, got {val!r}"
            )

    if policy == "ltm":
        require_inputs(
            {
                "last_full_fiscal_year": periods.last_full_fiscal_year,
                "current_year_to_date": periods.current_year_to_date,
                "prior_year_to_date": periods.prior_year_to_date,
            },
            ("last_full_fiscal_year", "current_year_to_date", "prior_year_to_date"),
            model,
        )
        value = (
            periods.last_full_fiscal_year
            + periods.current_year_to_date
            - periods.prior_year_to_date
        )

View on GitHub (pinned to 80ffdda44c)

Solutions

  1. Find which period value is non-finite: log the peer name and the four period values before calling run_comps.
  2. Drop or repair peers with NaN/inf periods (fillna with sourced estimates or exclude the peer).
  3. Guard your ingestion layer: math.isfinite check per numeric field before constructing fiscal periods.
  4. If NaN means 'not reported', pass None instead so calendarise_metric treats it as missing rather than invalid.

Example fix

# before
periods = {"last_full_fiscal_year": df.iloc[0]["fy_revenue"], ...}  # may be NaN

# after
import math
periods = {k: (v if v is None or math.isfinite(v) else None) for k, v in raw_periods.items()}
Defensive patterns

Strategy: validation

Validate before calling

import math
for name, v in periods.items():
    if v is not None and not math.isfinite(v):
        raise ValueError(f'non-finite period {name}: {v!r}')

Type guard

def has_finite_periods(p):
    return all(v is None or (isinstance(v, (int, float)) and math.isfinite(v)) for v in p.values())

Try / catch

from quantlib.valuation.contracts import ValuationError
try:
    run_comps(target, peers, policy)
except ValuationError as e:
    if 'must be a finite number' in str(e):
        log.warning('dropping peer with non-finite periods: %s', e)
    raise

Prevention

When it happens

Trigger: Calling run_comps / peer_multiple_set where a peer or target's fiscal periods dict contains math.nan, float('inf'), or a numpy NaN for any period key; typically the data came from a pandas frame or JSON with missing values coerced to NaN.

Common situations: Loading peer financials from CSV/DataFrame where missing cells become NaN; joining data sources that emit nulls; upstream arithmetic producing inf (division by zero) that is fed straight into fiscal periods.

Related errors

AI-assisted analysis of HKUDS/Vibe-Trading@80ffdda44c (2026-08-28). Data as JSON: /api/errors/10f7ebb7e6c13229. Report an issue: GitHub.