Bug/sc 458619/allow negative values on most common approx

raster_loader/io/common.py

@@ -185,7 +185,6 @@ def get_color_table(raster_dataset: rasterio.io.DatasetReader, band: int):
         return None
 
 
-
 def rasterio_metadata(
     file_path: str,
     bands_info: List[Tuple[int, str]],
@@ -424,7 +423,12 @@ 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
 
@@ -451,10 +455,22 @@ 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:
@@ -495,8 +511,14 @@ 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