fix: add src/evalml/leadtimes.py, missing from #272

src/evalml/leadtimes.py

@@ -0,0 +1,62 @@
+"""Lead times each participant (run or baseline) produces, per parameter."""
+
+import logging
+import re
+
+PARAMS_WITHOUT_STEP_ZERO_VALUE = {
+    "TOT_PREC",
+    "tp",
+    "CLCT",
+    "tcc",
+    "CLCL",
+    "lcc",
+    "CLCM",
+    "mcc",
+    "CLCH",
+    "hcc",
+    "SSRD",
+    "ssrd",
+}
+
+
+def resolve_leadtimes(steps_spec, requested="all", param=None):
+    """Lead times to compute for a single participant.
+
+    A run or baseline produces only the lead times in its own ``steps`` spec
+    (``start/stop/step``, hours). This returns those of the ``requested``
+    selection that the participant actually produces — the literal ``"all"``
+    (every produced lead time) or an explicit list of ints — so a 36h lead is
+    never requested of an ICON-CH1 baseline (steps ``0/33/6``), nor a >120h
+    lead of ICON-CH2. Explicitly requested lead times the participant cannot
+    produce are skipped with a warning. For accumulated ``param``s, lead times
+    shorter than one step spacing are dropped (no accumulation window).
+    """
+    start, end, step = map(int, steps_spec.split("/"))
+    supported = set(range(start, end + 1, step))
+    wanted = supported if requested == "all" else set(requested)
+
+    unsupported = sorted(wanted - supported)
+    if unsupported:
+        logging.getLogger("snakemake").warning(
+            "Skipping lead time(s) %sh: not produced by forecast steps '%s'.",
+            unsupported,
+            steps_spec,
+        )
+
+    valid = wanted & supported
+    # Showcase-facing names encode an accumulation window as a numeric suffix
+    # (e.g. "TOT_PREC1", "TOT_PREC6"); strip it before checking membership so
+    # those aggregated variants are recognized the same as their base param.
+    base_param = re.sub(r"\d+$", "", param) if param else param
+    if (
+        param in PARAMS_WITHOUT_STEP_ZERO_VALUE
+        or base_param in PARAMS_WITHOUT_STEP_ZERO_VALUE
+    ):
+        # For aggregated params (e.g. TOT_PREC6), the accumulation window encoded
+        # in the suffix is the minimum valid lead time — not the run's step spacing.
+        # At shorter leads the de-accumulation window doesn't fit (prev_lt < 0),
+        # so the frame would fall back to showing a raw cumulative total.
+        agg_match = re.search(r"\d+$", param) if param and base_param != param else None
+        min_lt = int(agg_match.group()) if agg_match else step
+        valid = {lt for lt in valid if lt >= min_lt}
+    return sorted(valid)