ShoreDataLabs performance improvements

rounds/1_histogram/solution.py

@@ -5,10 +5,20 @@
 own faster implementation.
 """
 
+import numpy as np
+
 
 def compute_histogram(path: str) -> dict[bytes, int]:
     """Frequency of every 2-byte bigram in the file at ``path``."""
-    # TODO: remove this delegation and write your own implementation here.
-    from .baseline import compute_histogram as _baseline
+    with open(path, "rb") as f:
+        data = f.read()
+
+    arr = np.frombuffer(data, dtype=np.uint8)
+    # Encode each bigram as a uint16 index: high_byte * 256 + low_byte
+    indices = arr[:-1].astype(np.uint16) * 256 + arr[1:]
+    counts = np.bincount(indices, minlength=65536)
 
-    return _baseline(path)
+    result: dict[bytes, int] = {}
+    for idx in np.nonzero(counts)[0]:
+        result[bytes([idx >> 8, idx & 0xFF])] = int(counts[idx])
+    return result

rounds/2_corruption/solution.py

@@ -5,10 +5,27 @@
 own faster implementation.
 """
 
-from .baseline import find_corruptions as _baseline
+import numpy as np
 
 
 def find_corruptions(ref_path: str, cor_path: str) -> list[tuple[int, int]]:
     """Return ``[(offset, length), ...]`` for every differing byte range."""
-    # TODO: remove this delegation and write your own implementation here.
-    return _baseline(ref_path, cor_path)
+    ref = np.fromfile(ref_path, dtype=np.uint8)
+    cor = np.fromfile(cor_path, dtype=np.uint8)
+
+    if len(ref) != len(cor):
+        raise ValueError("reference and corrupted files differ in length")
+
+    # Single vectorised comparison — runs entirely in C.
+    diff_indices = np.where(ref != cor)[0]
+
+    if len(diff_indices) == 0:
+        return []
+
+    # Find the boundaries between consecutive runs.
+    # A new run starts wherever the gap between adjacent indices exceeds 1.
+    gaps = np.where(np.diff(diff_indices) > 1)[0]
+    starts = diff_indices[np.concatenate(([0], gaps + 1))]
+    ends = diff_indices[np.concatenate((gaps, [len(diff_indices) - 1]))]
+
+    return [(int(s), int(e - s + 1)) for s, e in zip(starts, ends)]