Merge pull request #1 from adriencaccia/perf/adriencaccia

perf improvements for adriencaccia

Author: adriencaccia

.python-version

@@ -1 +1 @@
-3.15t
+3.15

README.md

@@ -1,4 +1,4 @@
-# Python Performance Lab: Sharpening Your Instincts
+# Python Performance Lab: Sharpening Your Instincts - `adriencaccia`
 
 A PyCon US 2026 hands-on tutorial. You optimize intentionally slow Python code
 across three rounds plus a team challenge, measuring every change with

rounds/1_histogram/solution.py

@@ -8,7 +8,21 @@
 
 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
+    # Step 1: read the whole file into memory as a single bytes object.
+    with open(path, "rb") as f:
+        data = f.read()
 
-    return _baseline(path)
+    # Create a 2D matrix to count bigrams
+    counts = [[0] * 256 for _ in range(256)]
+
+    for i in range(len(data) - 1):
+        # Increment the count in each cell
+        counts[data[i]][data[i + 1]] += 1
+
+    # Convert the matrix to the original format
+    output = {}
+    for i in range(256):
+        for j in range(256):
+            if counts[i][j] > 0:
+                output[bytes([i, j])] = counts[i][j]
+    return output

rounds/3_dna/solution.py

@@ -5,13 +5,80 @@
 own faster implementation.
 """
 
-from .baseline import find_matches as _baseline
+from __future__ import annotations
+
+import os
+from concurrent.futures import ThreadPoolExecutor
+
+_NL = 0x0A  # b"\n"
 
 
 def find_matches(fasta_path: str, pattern: bytes) -> list[tuple[str, list[int]]]:
-    """Find every FASTA record whose sequence contains ``pattern``.
+    with open(fasta_path, "rb") as f:
+        data = f.read()
+
+    # Step 1: locate every record start. A record starts with ``>`` either at
+    # offset 0 or immediately after a ``\n``.
+    starts: list[int] = []
+    i = 0
+    while True:
+        p = data.find(b">", i)
+        if p == -1:
+            break
+        if p == 0 or data[p - 1] == _NL:
+            starts.append(p)
+        i = p + 1
+    starts.append(len(data))  # sentinel marking the end of the last record.
+
+    num_records = len(starts) - 1
+    if num_records <= 0:
+        return []
+
+    # Step 2: parallel scan. Choose enough batches to keep workers balanced
+    # even when record sizes vary.
+    n_workers = max(1, os.cpu_count() or 1)
+    batches = max(1, n_workers * 4)
+    batch_size = max(1, (num_records + batches - 1) // batches)
+
+    def scan_batch(start_idx: int, end_idx: int) -> list[tuple[int, str, list[int]]]:
+        out: list[tuple[int, str, list[int]]] = []
+        for j in range(start_idx, end_idx):
+            rec_start = starts[j]
+            rec_end = starts[j + 1]
+
+            # Locate the end of the header line within this record's slice.
+            nl = data.find(b"\n", rec_start, rec_end)
+            if nl <= rec_start:
+                continue  # Malformed or header-only.
+
+            record_id = data[rec_start + 1 : nl].decode("ascii").strip()
+
+            # Contiguous sequence: drop the newlines so matches that straddle
+            # line breaks are still found by ``bytes.find``.
+            sequence = data[nl + 1 : rec_end].replace(b"\n", b"")
+
+            positions: list[int] = []
+            s = 0
+            while True:
+                p = sequence.find(pattern, s)
+                if p == -1:
+                    break
+                positions.append(p)
+                s = p + 1
+
+            if positions:
+                out.append((j, record_id, positions))
+        return out
+
+    with ThreadPoolExecutor(max_workers=n_workers) as pool:
+        futures = [
+            pool.submit(scan_batch, lo, min(lo + batch_size, num_records))
+            for lo in range(0, num_records, batch_size)
+        ]
+        chunks = [f.result() for f in futures]
 
-    Returns ``[(record_id, [positions...]), ...]`` in file order.
-    """
-    # TODO: remove this delegation and write your own implementation here.
-    return _baseline(fasta_path, pattern)
+    # Step 3: flatten and restore file order (record index is monotonic per
+    # batch, but batches finish in arbitrary order).
+    flat = [item for chunk in chunks for item in chunk]
+    flat.sort(key=lambda triple: triple[0])
+    return [(rid, positions) for _, rid, positions in flat]