fix(data): re-derive early_january_2023_weather from weather (real temp/dewp/humid/pressure)

tests/test_data_and_theory.py

@@ -40,6 +40,18 @@ def test_unknown_dataset_raises():
         md.load_dataset("does_not_exist")
 
 
+def test_early_january_2023_weather_has_real_measurements():
+    # Regression: the R dataset ships temp/dewp/humid/pressure as all-NA; this
+    # package derives the table from `weather`, so those columns must be numeric
+    # and fully populated (Newark, first 15 days of January 2023 = 360 hours).
+    df = md.load_early_january_2023_weather()
+    assert df.height == 360
+    assert df["origin"].unique().to_list() == ["EWR"]
+    for col in ("temp", "dewp", "humid", "pressure"):
+        assert df.schema[col].is_numeric()
+        assert df[col].null_count() == 0
+
+
 def test_t_confidence_interval_matches_manual():
     x = np.array([2.0, 4.0, 4.0, 4.0, 5.0, 5.0, 7.0, 9.0])
     out = theory.t_confidence_interval(x, level=0.95)

tools/build_data.py

@@ -44,7 +44,6 @@ def _write(df: pl.DataFrame, name: str) -> None:
 def main() -> None:
     from_r = [
         "envoy_flights",
-        "early_january_2023_weather",
         "weather",
         "flights",
         "airlines",
@@ -82,6 +81,15 @@ def main() -> None:
     for name in from_r:
         _write(_read_csv(name), name)
 
+    # early_january_2023_weather: the R `moderndive` dataset ships with
+    # temp/dewp/humid/pressure entirely NA, so derive it from `weather` (Newark,
+    # first 15 days of January 2023) to get the real, correctly-typed values.
+    weather = _read_csv("weather")
+    ejw = weather.filter(
+        (pl.col("origin") == "EWR") & (pl.col("month") == 1) & (pl.col("day") <= 15)
+    ).sort("time_hour")
+    _write(ejw, "early_january_2023_weather")
+
     # Gapminder ships a CSV inside the `gapminder` PyPI package. We read that file
     # directly (the package's Python import is broken on 3.14: it uses the removed
     # pkg_resources). The data matches the canonical R gapminder dataset.