Bug/sc 458619/allow negative values on most common approx (#159)

Co-authored-by: cayetanobv <[email protected]>

Author: rantolin

CHANGELOG.md

@@ -7,6 +7,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 <!-- insertion marker -->
 
+## [0.10.2] 2025-01-14
+
+<small>[Compare with latest](https://github.com/CartoDB/raster-loader/compare/v0.10.1...HEAD)</small>
+
+### Fixed
+
+- Fix: OverflowError error when casting approx sum to integer ([46cab53](https://github.com/CartoDB/raster-loader/commit/46cab53bbf71a86a7df784922956eb03f9dbb327) by Roberto Antolín).
+- Fix: Compute approximate most common negative values ([f9f5ff5](https://github.com/CartoDB/raster-loader/commit/f9f5ff5010b1aea0d13afbea6d1869d4094fa7d7) by Roberto Antolín).
+
 ## [0.10.1] 2025-01-13
 
 <small>[Compare with latest](https://github.com/CartoDB/raster-loader/compare/57d55999704fb003da2947db65d5617e27c5c104...HEAD)</small>

raster_loader/io/common.py

@@ -447,7 +447,12 @@ def not_enough_samples():
             )
             if not raster_is_masked:
                 for band in bands:
-                    not_masked_samples[band].append(sample[band - 1])
+                    band_sample = sample[band - 1]
+                    is_valid_sample = not (
+                        np.isinf(band_sample) or np.isnan(band_sample)
+                    )
+                    if is_valid_sample:
+                        not_masked_samples[band].append(band_sample)
 
         iterations += 1
 
@@ -474,10 +479,22 @@ def not_enough_samples():
 
 def most_common_approx(samples: List[Union[int, float]]) -> Dict[int, int]:
     """Compute the most common values in a list of int samples."""
-    counts = np.bincount(samples)
+    print("Computing most common values...")
+
+    samples_array = np.array(samples)
+    min_val = int(np.floor(samples_array.min()))
+    max_val = int(np.ceil(samples_array.max()))
+
+    # +2 allows to include max_val in the last bin
+    bins = np.arange(min_val, max_val + 2)
+
+    counts, bin_edges = np.histogram(samples_array, bins=bins)
+
     nth = min(DEFAULT_MAX_MOST_COMMON, len(counts))
+    counts = np.bincount(samples)
     idx = np.argpartition(counts, -nth)[-nth:]
-    return dict([(int(i), int(counts[i])) for i in idx if counts[i] > 0])
+
+    return {int(bin_edges[i]): int(counts[i]) for i in idx if counts[i] > 0}
 
 
 def compute_quantiles(data: List[Union[int, float]], cast_function: Callable) -> dict:
@@ -518,8 +535,14 @@ def raster_band_approx_stats(
     _sum = 0
     sum_squares = 0
     if count > 0:
-        _sum = int(np.sum(samples_band))
-        sum_squares = int(np.sum(np.array(samples_band) ** 2))
+        try:
+            _sum = int(np.sum(samples_band))
+        except (OverflowError, ValueError):
+            _sum = 0
+        try:
+            sum_squares = int(np.sum(np.array(samples_band) ** 2))
+        except (OverflowError, ValueError):
+            sum_squares = 0
 
     if basic_stats:
         quantiles = None