Analysis Page Data and Text Corrections

backend/app/schemas/analysis.py

@@ -27,8 +27,9 @@ class TrendLine(BaseModel):
     slope: float
     intercept: float
     r_squared: float
-    pearson_r: float
-    spearman_rho: float
+    # None when undefined: fewer than 3 points or a constant axis.
+    pearson_r: float | None
+    spearman_rho: float | None
     confidence_upper: list[ConfidenceBandPoint]
     confidence_lower: list[ConfidenceBandPoint]
 
@@ -48,6 +49,9 @@ class CorrelationResponse(BaseModel):
     # and x/y stats are still computed over the full set. Equal otherwise.
     total_count: int = 0
     sampled_count: int = 0
+    # True when a pixel-domain metric (hfr) is involved and the filtered frames
+    # span more than one plate scale, so raw pixel values are not comparable.
+    mixed_plate_scales: bool = False
 
 
 class HistogramBin(BaseModel):
@@ -61,6 +65,9 @@ class DistributionResponse(BaseModel):
     stats: SummaryStats
     metric: str
     skewness: float
+    # True when a pixel-domain metric (hfr) is involved and the filtered frames
+    # span more than one plate scale, so raw pixel values are not comparable.
+    mixed_plate_scales: bool = False
 
 
 class BoxPlotGroup(BaseModel):
@@ -74,10 +81,20 @@ class BoxPlotGroup(BaseModel):
     count: int
 
 
+class SkippedGroup(BaseModel):
+    group_name: str
+    count: int
+
+
 class BoxPlotResponse(BaseModel):
     groups: list[BoxPlotGroup]
     metric: str
     group_by: str
+    # Groups dropped for having fewer than 4 values, sorted by group_name.
+    skipped_groups: list[SkippedGroup] = []
+    # True when a pixel-domain metric (hfr) is involved and the filtered frames
+    # span more than one plate scale, so raw pixel values are not comparable.
+    mixed_plate_scales: bool = False
 
 
 class TimeSeriesPoint(BaseModel):
@@ -98,6 +115,9 @@ class TimeSeriesResponse(BaseModel):
     ma_30: list[MovingAveragePoint]
     metric: str
     month_boundaries: list[str]
+    # True when a pixel-domain metric (hfr) is involved and the filtered frames
+    # span more than one plate scale, so raw pixel values are not comparable.
+    mixed_plate_scales: bool = False
 
 
 class MatrixCell(BaseModel):
@@ -111,6 +131,9 @@ class MatrixResponse(BaseModel):
     cells: list[MatrixCell]
     x_metrics: list[str]
     y_metrics: list[str]
+    # True when a pixel-domain metric (hfr) is involved and the filtered frames
+    # span more than one plate scale, so raw pixel values are not comparable.
+    mixed_plate_scales: bool = False
 
 
 class CompareGroupStats(BaseModel):

backend/tests/test_analysis_stats.py

@@ -0,0 +1,85 @@
+"""Pure-logic tests for the analysis statistics helpers (issue #311)."""
+import statistics
+
+import pytest
+
+from app.api.analysis import (
+    _histogram_bins,
+    _iqr_fences,
+    _join_target_names,
+    _pct_lower,
+    _pearson_r,
+    _spearman_rho,
+    _t_crit_95,
+)
+
+
+def test_histogram_counts_sum_to_n_and_keep_the_maximum():
+    # v_min + 10 * ((v_max - v_min) / 10) lands below v_max in floating point.
+    values = [0.0, 2.6245] + [0.1 * i for i in range(1, 26)]
+    bins = _histogram_bins(values, 10)
+    assert len(bins) == 10
+    assert sum(count for _, _, count in bins) == len(values)
+    assert bins[-1][1] == 2.6245
+    assert bins[-1][2] >= 1
+
+
+def test_histogram_constant_values():
+    bins = _histogram_bins([3.0, 3.0, 3.0], 3)
+    assert sum(count for _, _, count in bins) == 3
+
+
+def test_pct_lower_uses_the_larger_median_and_is_symmetric():
+    assert round(_pct_lower(1.772, 3.997)) == 56
+    assert _pct_lower(1.772, 3.997) == _pct_lower(3.997, 1.772)
+    assert _pct_lower(0.0, 0.0) == 0.0
+    assert _pct_lower(0.0, 2.0) == 100.0
+
+
+def test_pearson_r_none_when_undefined():
+    assert _pearson_r([1.0, 2.0], [1.0, 2.0]) is None
+    assert _pearson_r([1.0, 1.0, 1.0, 1.0], [1.0, 2.0, 3.0, 4.0]) is None
+    assert _pearson_r([1.0, 2.0, 3.0, 4.0], [5.0, 5.0, 5.0, 5.0]) is None
+    assert _spearman_rho([1.0, 2.0], [1.0, 2.0]) is None
+    assert _spearman_rho([1.0, 1.0, 1.0], [1.0, 2.0, 3.0]) is None
+
+
+def test_pearson_r_perfect_correlation():
+    xs = [1.0, 2.0, 3.0, 4.0, 5.0]
+    assert _pearson_r(xs, [2 * x + 1 for x in xs]) == pytest.approx(1.0)
+    assert _pearson_r(xs, [-x for x in xs]) == pytest.approx(-1.0)
+
+
+def test_t_critical_values():
+    assert _t_crit_95(2) == pytest.approx(4.303, abs=0.001)
+    assert _t_crit_95(30) == pytest.approx(2.042, abs=0.001)
+    assert _t_crit_95(31) == 1.96
+
+
+def _brute_force_outlier(x, y, xs, ys):
+    for vals, val in [(xs, x), (ys, y)]:
+        s = sorted(vals)
+        n = len(s)
+        q1 = statistics.median(s[: n // 2])
+        q3 = statistics.median(s[(n + 1) // 2 :])
+        iqr = q3 - q1
+        if val < q1 - 1.5 * iqr or val > q3 + 1.5 * iqr:
+            return True
+    return False
+
+
+def test_iqr_fences_match_per_point_computation():
+    xs = [1.0, 1.2, 0.9, 1.1, 1.05, 9.0, 1.15, 0.95, -6.0]
+    ys = [2.0, 2.1, 1.9, 40.0, 2.05, 2.0, 1.95, 2.2, 2.1]
+    x_lo, x_hi = _iqr_fences(xs)
+    y_lo, y_hi = _iqr_fences(ys)
+    flags = [x < x_lo or x > x_hi or y < y_lo or y > y_hi for x, y in zip(xs, ys)]
+    assert flags == [_brute_force_outlier(x, y, xs, ys) for x, y in zip(xs, ys)]
+    assert flags.count(True) == 3
+
+
+def test_join_target_names():
+    assert _join_target_names([]) is None
+    assert _join_target_names(["M 31"]) == "M 31"
+    assert _join_target_names(["NGC 7000", "M 31"]) == "M 31, NGC 7000"
+    assert _join_target_names(["d", "b", "a", "c", "e"]) == "a, b, c +2 more"