Side by side animations

config/interpolators-ich1.yaml

@@ -23,6 +23,7 @@ runs:
         checkpoint: https://service.meteoswiss.ch/mlstore#/experiments/602/runs/c30490b6ba064e4db03b430f3a2595ad
         config: resources/inference/configs/sgm-multidataset-forecaster-global-ich1-oper.yaml
         steps: 0/120/6
+        label: stage_E
         extra_requirements:
           - git+https://github.com/ecmwf/anemoi-inference.git@e369b1a90313e9701db13f63364a467aa281cf36
       extra_requirements:

config/showcase-interpolators-ich1.yaml

@@ -0,0 +1,27 @@
+# Showcase config for interpolators-ich1.
+# Includes the base config and overrides only the showcase section.
+#
+#   .venv/bin/evalml showcase config/showcase-interpolators-ich1.yaml
+
+include: interpolators-ich1.yaml
+
+showcase:
+  params:
+    - T_2M
+    - SP_10M
+    - TOT_PREC
+  meteograms:
+    enabled: false
+    stations: [JUN]
+  animations:
+    enabled: true
+    domains:
+      - europe
+      - switzerland
+      - globe
+    speed: 10  # simulated hours per second
+    runs:
+      - Varda-Single
+    comparisons:
+      - left: Varda-Single
+        right: KENDA-CH1

src/data_input/__init__.py

@@ -64,7 +64,10 @@ def load_analysis_data_from_zarr(
         "PMSL": "msl",
         "TOT_PREC": "tp",
     }
-    tot_prec_string = "TOT_PREC_6H" if min(np.diff(steps)) == 6 else "TOT_PREC_1H"
+    _diffs = np.diff(steps)
+    tot_prec_string = (
+        "TOT_PREC_6H" if len(_diffs) > 0 and min(_diffs) == 6 else "TOT_PREC_1H"
+    )
     PARAMS_MAP_COSMO1 = {
         v: v.replace("TOT_PREC", tot_prec_string) for v in PARAMS_MAP_COSMO2.keys()
     }

src/evalml/cli.py

@@ -69,9 +69,24 @@ def generate_graph(
     click.echo(f"Graph saved to {output_file}")
 
 
+def _deep_merge(base: dict, override: dict) -> dict:
+    """Recursively merge override into base. Override wins on conflicts."""
+    result = dict(base)
+    for key, value in override.items():
+        if key in result and isinstance(result[key], dict) and isinstance(value, dict):
+            result[key] = _deep_merge(result[key], value)
+        else:
+            result[key] = value
+    return result
+
+
 def load_yaml(path: Path) -> dict[str, Any]:
     with path.open("r") as f:
-        return yaml.safe_load(f)
+        data = yaml.safe_load(f)
+    if include := data.pop("include", None):
+        base = load_yaml(path.parent / include)
+        data = _deep_merge(base, data)
+    return data
 
 
 def workflow_options(func):

src/evalml/config.py

@@ -1,5 +1,5 @@
 from pathlib import Path
-from typing import Dict, List, Any, ClassVar, FrozenSet
+from typing import Dict, List, Any, ClassVar, FrozenSet, Optional
 
 from pydantic import BaseModel, Field, RootModel, field_validator
 
@@ -212,6 +212,12 @@ class BaselineItem(BaseModel):
     baseline: BaselineConfig
 
 
+class RegionConfig(BaseModel):
+    """A custom map region defined by name, extent, and projection."""
+
+    name: str = Field(..., description="Name for the custom region (used as wildcard).")
+
+
 class DomainConfig(BaseModel):
     """A custom map domain defined by name, extent, and projection."""
 
@@ -228,6 +234,13 @@ class DomainConfig(BaseModel):
     model_config = {"extra": "forbid"}
 
 
+class AnimationComparison(BaseModel):
+    """A side-by-side comparison animation between two runs."""
+
+    left: str = Field(..., description="Label of the run shown in the left panel.")
+    right: str = Field(..., description="Label of the run shown in the right panel.")
+
+
 class MeteogramConfig(BaseModel):
     """Configuration for meteogram generation."""
 
@@ -248,7 +261,7 @@ class AnimationsConfig(BaseModel):
         default=True,
         description="Whether to generate forecast animations (GIFs per param and region).",
     )
-    domains: List[str | DomainConfig] = Field(
+    domains: List[str | RegionConfig] = Field(
         default=["globe", "europe", "switzerland"],
         description=(
             "Domains to generate animations for. Each entry is either a named domain "
@@ -257,6 +270,25 @@ class AnimationsConfig(BaseModel):
             "[lon_min, lon_max, lat_min, lat_max], and optional 'projection'."
         ),
     )
+    speed: float = Field(
+        default=10.0,
+        gt=0,
+        description="Animation playback speed in simulated hours per second.",
+    )
+    runs: Optional[List[str]] = Field(
+        default=None,
+        description=(
+            "Labels of runs to generate individual animations for. "
+            "Defaults to all candidate runs when omitted."
+        ),
+    )
+    comparisons: List[AnimationComparison] = Field(
+        default=[],
+        description=(
+            "Side-by-side two-panel comparison animations. Each entry specifies "
+            "the labels of the left and right panel runs."
+        ),
+    )
 
 
 class ShowcaseConfig(BaseModel):

src/plotting/compat.py

@@ -1,4 +1,4 @@
-from datetime import datetime
+from datetime import datetime, timedelta
 from pathlib import Path
 
 import earthkit.data as ekd
@@ -77,6 +77,60 @@ def load_state_from_grib(
     return state
 
 
+def load_state_from_zarr(
+    zarr_root: Path,
+    reftime: datetime,
+    lead_time_hours: int,
+    params: list[str],
+    source_type: str = "analysis",
+) -> dict:
+    """Load a single time step from a zarr source into the state dict used by StatePlotter.
+
+    Parameters
+    ----------
+    zarr_root:
+        Path to the zarr dataset.
+    reftime:
+        Forecast reference time (init time).
+    lead_time_hours:
+        Lead time in hours to load.
+    params:
+        List of parameter names (ICON convention, e.g. ``['U_10M', 'V_10M']``).
+    source_type:
+        ``'analysis'`` for truth zarrs (loads via ``load_analysis_data_from_zarr``),
+        ``'baseline'`` for baseline forecast zarrs.
+    """
+    from data_input import load_analysis_data_from_zarr, load_baseline_from_zarr
+
+    steps = [lead_time_hours]
+
+    if source_type == "analysis":
+        ds = load_analysis_data_from_zarr(zarr_root, reftime, steps, params)
+        ds_t = ds.isel(time=0) if "time" in ds.dims else ds.squeeze()
+    else:
+        ds = load_baseline_from_zarr(zarr_root, reftime, steps, params)
+        ds_t = ds.isel(lead_time=0) if "lead_time" in ds.dims else ds.squeeze()
+
+    lat = ds_t.lat.values.flatten()
+    lon = ds_t.lon.values.flatten()
+
+    hull = MultiPoint(list(zip(lon.tolist(), lat.tolist()))).convex_hull
+    state = {
+        "forecast_reference_time": reftime,
+        "valid_time": reftime + timedelta(hours=lead_time_hours),
+        "longitudes": lon,
+        "latitudes": lat,
+        "lam_envelope": gpd.GeoSeries([hull], crs="EPSG:4326"),
+        "fields": {},
+    }
+    for param in params:
+        if param in ds_t.data_vars:
+            state["fields"][param] = ds_t[param].values.flatten()
+        else:
+            state["fields"][param] = np.full(lat.size, np.nan, dtype=float)
+    return state
+
+
 def load_state_from_raw(
     file: Path, paramlist: list[str] | None = None
 ) -> dict[str, np.ndarray | dict[str, np.ndarray]]:

tests/unit/test_config.py

@@ -1,8 +1,47 @@
 import pytest
 
+from os.path import Path
+from evalml.cli import _deep_merge, load_yaml
 from evalml.config import ConfigModel
 
 
+def test_deep_merge_override_wins():
+    base = {"a": 1, "b": {"x": 1, "y": 2}}
+    override = {"b": {"y": 99}, "c": 3}
+    result = _deep_merge(base, override)
+    assert result == {"a": 1, "b": {"x": 1, "y": 99}, "c": 3}
+
+
+def test_deep_merge_non_dict_override_replaces():
+    base = {"a": {"x": 1}}
+    override = {"a": [1, 2, 3]}
+    result = _deep_merge(base, override)
+    assert result["a"] == [1, 2, 3]
+
+
+def test_load_yaml_without_include(tmp_path):
+    f = tmp_path / "config.yaml"
+    f.write_text("a: 1\n")
+    assert load_yaml(f) == {"a": 1}
+
+
+def test_load_yaml_include_merges_base(tmp_path):
+    base = tmp_path / "base.yaml"
+    base.write_text("a: 1\nb:\n  x: 1\n  y: 2\n")
+
+    child = tmp_path / "child.yaml"
+    child.write_text("include: base.yaml\nb:\n  y: 99\nc: 3\n")
+
+    result = load_yaml(child)
+    assert result == {"a": 1, "b": {"x": 1, "y": 99}, "c": 3}
+
+
+def test_load_yaml_include_validates_as_config_model():
+    path = Path("config/showcase-interpolators-ich1.yaml")
+    data = load_yaml(path)
+    _ = ConfigModel.model_validate(data)
+
+
 def test_example_forecasters_config(example_forecasters_config):
     """Test that the example config loads correctly."""
 

workflow/Snakefile

@@ -140,14 +140,26 @@ rule showcase_all:
             expand(
                 rules.make_forecast_animation.output,
                 init_time=[t.strftime("%Y%m%d%H%M") for t in REFTIMES],
-                run_id=CANDIDATES,
+                run_id=SHOWCASE_ANIMATION_RUN_IDS,
                 param=SHOWCASE_PARAMS,
                 region=list(SHOWCASE_REGIONS.keys()),
                 showcase=EXPERIMENT_NAME,
             )
             if config["showcase"]["animations"]["enabled"]
             else []
         ),
+        (
+            expand(
+                rules.make_comparison_animation.output,
+                init_time=[t.strftime("%Y%m%d%H%M") for t in REFTIMES],
+                comparison_id=[c["id"] for c in SHOWCASE_COMPARISONS],
+                param=SHOWCASE_PARAMS,
+                region=list(SHOWCASE_REGIONS.keys()),
+                showcase=EXPERIMENT_NAME,
+            )
+            if config["showcase"]["animations"]["enabled"] and SHOWCASE_COMPARISONS
+            else []
+        ),
         (
             expand(
                 rules.plot_meteogram.output,