[Autoloop: perf-comparison]

benchmarks/pandas/bench_cum_ops.py

@@ -0,0 +1,65 @@
+"""
+Benchmark: Series.cumsum / cumprod / cummax / cummin / DataFrame.cumsum
+Mirrors tsb bench_cum_ops.ts
+"""
+import json
+import time
+import pandas as pd
+
+N = 100_000
+WARMUP = 5
+ITERS = 20
+
+# Matching data
+data = [(i % 100) + 1 for i in range(N)]
+series = pd.Series(data, dtype=float)
+
+col1 = [(i % 100) + 1 for i in range(N)]
+col2 = [((i * 3) % 100) + 1 for i in range(N)]
+df = pd.DataFrame({"a": col1, "b": col2}, dtype=float)
+
+# Warm-up
+for _ in range(WARMUP):
+    series.cumsum()
+    series.cummax()
+    df.cumsum()
+
+# Measured: cumsum
+t0 = time.perf_counter()
+for _ in range(ITERS):
+    series.cumsum()
+total_cumsum = (time.perf_counter() - t0) * 1000
+
+# Measured: cumprod
+t0 = time.perf_counter()
+for _ in range(ITERS):
+    series.cumprod()
+total_cumprod = (time.perf_counter() - t0) * 1000
+
+# Measured: cummax
+t0 = time.perf_counter()
+for _ in range(ITERS):
+    series.cummax()
+total_cummax = (time.perf_counter() - t0) * 1000
+
+# Measured: cummin
+t0 = time.perf_counter()
+for _ in range(ITERS):
+    series.cummin()
+total_cummin = (time.perf_counter() - t0) * 1000
+
+# Measured: DataFrame.cumsum
+t0 = time.perf_counter()
+for _ in range(ITERS):
+    df.cumsum()
+total_df = (time.perf_counter() - t0) * 1000
+
+total_ms = total_cumsum + total_cumprod + total_cummax + total_cummin + total_df
+mean_ms = total_ms / (ITERS * 5)
+
+print(json.dumps({
+    "function": "cum_ops",
+    "mean_ms": round(mean_ms, 4),
+    "iterations": ITERS * 5,
+    "total_ms": round(total_ms, 4),
+}))

benchmarks/pandas/bench_replace.py

@@ -0,0 +1,47 @@
+"""
+Benchmark: Series.replace / DataFrame.replace
+Mirrors tsb bench_replace.ts
+"""
+import json
+import time
+import pandas as pd
+import numpy as np
+
+N = 100_000
+WARMUP = 5
+ITERS = 20
+
+# Build data matching the TypeScript benchmark
+data = [i % 10 for i in range(N)]
+series = pd.Series(data)
+
+col1 = [i % 10 for i in range(N)]
+col2 = [(i * 3) % 10 for i in range(N)]
+df = pd.DataFrame({"a": col1, "b": col2})
+
+# Warm-up
+for _ in range(WARMUP):
+    series.replace(5, 99)
+    df.replace(5, 99)
+
+# Measured: Series.replace scalar
+t0 = time.perf_counter()
+for i in range(ITERS):
+    series.replace(i % 10, 99)
+total_series = (time.perf_counter() - t0) * 1000
+
+# Measured: DataFrame.replace scalar
+t0 = time.perf_counter()
+for i in range(ITERS):
+    df.replace(i % 10, 99)
+total_df = (time.perf_counter() - t0) * 1000
+
+total_ms = total_series + total_df
+mean_ms = total_ms / (ITERS * 2)
+
+print(json.dumps({
+    "function": "replace",
+    "mean_ms": round(mean_ms, 4),
+    "iterations": ITERS * 2,
+    "total_ms": round(total_ms, 4),
+}))

benchmarks/tsb/bench_cum_ops.ts

@@ -0,0 +1,70 @@
+/**
+ * Benchmark: cumsum / cumprod / cummax / cummin (Series and DataFrame)
+ * Mirrors pandas Series.cumsum(), DataFrame.cumsum(), etc.
+ */
+import { Series, DataFrame } from "../../src/index.ts";
+import {
+  cumsum,
+  cumprod,
+  cummax,
+  cummin,
+  dataFrameCumsum,
+} from "../../src/stats/cum_ops.ts";
+
+const N = 100_000;
+
+// Numeric series for cumsum/cumprod/cummax/cummin
+const data = Array.from({ length: N }, (_, i) => (i % 100) + 1);
+const series = new Series({ data });
+
+// DataFrame with two columns
+const col1 = Array.from({ length: N }, (_, i) => (i % 100) + 1);
+const col2 = Array.from({ length: N }, (_, i) => ((i * 3) % 100) + 1);
+const df = DataFrame.fromColumns({ a: col1, b: col2 });
+
+const WARMUP = 5;
+const ITERS = 20;
+
+// --- warm-up ---
+for (let i = 0; i < WARMUP; i++) {
+  cumsum(series);
+  cummax(series);
+  dataFrameCumsum(df);
+}
+
+// --- measured: cumsum ---
+const t0cs = performance.now();
+for (let i = 0; i < ITERS; i++) cumsum(series);
+const totalCumsum = performance.now() - t0cs;
+
+// --- measured: cumprod ---
+const t0cp = performance.now();
+for (let i = 0; i < ITERS; i++) cumprod(series);
+const totalCumprod = performance.now() - t0cp;
+
+// --- measured: cummax ---
+const t0cx = performance.now();
+for (let i = 0; i < ITERS; i++) cummax(series);
+const totalCummax = performance.now() - t0cx;
+
+// --- measured: cummin ---
+const t0cn = performance.now();
+for (let i = 0; i < ITERS; i++) cummin(series);
+const totalCummin = performance.now() - t0cn;
+
+// --- measured: dataFrameCumsum ---
+const t0df = performance.now();
+for (let i = 0; i < ITERS; i++) dataFrameCumsum(df);
+const totalDf = performance.now() - t0df;
+
+const total_ms = totalCumsum + totalCumprod + totalCummax + totalCummin + totalDf;
+const mean_ms = total_ms / (ITERS * 5);
+
+console.log(
+  JSON.stringify({
+    function: "cum_ops",
+    mean_ms: parseFloat(mean_ms.toFixed(4)),
+    iterations: ITERS * 5,
+    total_ms: parseFloat(total_ms.toFixed(4)),
+  }),
+);

benchmarks/tsb/bench_replace.ts

@@ -0,0 +1,53 @@
+/**
+ * Benchmark: replaceSeries / replaceDataFrame
+ * Mirrors pandas Series.replace() and DataFrame.replace().
+ */
+import { Series, DataFrame } from "../../src/index.ts";
+import { replaceSeries, replaceDataFrame } from "../../src/stats/replace.ts";
+
+const N = 100_000;
+
+// Build a numeric series with values 0–9 (cycled) for scalar replace
+const data = Array.from({ length: N }, (_, i) => i % 10);
+const series = new Series({ data });
+
+// Build a DataFrame with two numeric columns
+const col1 = Array.from({ length: N }, (_, i) => i % 10);
+const col2 = Array.from({ length: N }, (_, i) => (i * 3) % 10);
+const df = DataFrame.fromColumns({ a: col1, b: col2 });
+
+const WARMUP = 5;
+const ITERS = 20;
+
+// --- warm-up ---
+for (let i = 0; i < WARMUP; i++) {
+  replaceSeries(series, { toReplace: 5, value: 99 });
+  replaceDataFrame(df, { toReplace: 5, value: 99 });
+}
+
+// --- measured: replaceSeries scalar ---
+const t0s = performance.now();
+for (let i = 0; i < ITERS; i++) {
+  replaceSeries(series, { toReplace: i % 10, value: 99 });
+}
+const totalSeries = performance.now() - t0s;
+
+// --- measured: replaceDataFrame scalar ---
+const t0d = performance.now();
+for (let i = 0; i < ITERS; i++) {
+  replaceDataFrame(df, { toReplace: i % 10, value: 99 });
+}
+const totalDf = performance.now() - t0d;
+
+// Report the average of the two operations
+const total_ms = totalSeries + totalDf;
+const mean_ms = total_ms / (ITERS * 2);
+
+console.log(
+  JSON.stringify({
+    function: "replace",
+    mean_ms: parseFloat(mean_ms.toFixed(4)),
+    iterations: ITERS * 2,
+    total_ms: parseFloat(total_ms.toFixed(4)),
+  }),
+);