Merge pull request #128 from githubnext/autoloop/perf-comparison-12683ddd005469f2

[Autoloop] [Autoloop: perf-comparison]

Author: mrjf

benchmarks/pandas/bench_concat.py

@@ -0,0 +1,28 @@
+"""Benchmark: concat — concatenate two 50k-row DataFrames"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 50_000
+WARMUP = 5
+ITERATIONS = 20
+
+vals1 = np.arange(ROWS, dtype=np.float64)
+vals2 = np.arange(ROWS, dtype=np.float64) * 2.0
+df1 = pd.DataFrame({"value": vals1})
+df2 = pd.DataFrame({"value": vals2})
+
+for _ in range(WARMUP):
+    pd.concat([df1, df2], ignore_index=True)
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    pd.concat([df1, df2], ignore_index=True)
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "concat",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_apply.py

@@ -0,0 +1,27 @@
+"""Benchmark: dataframe_apply — apply a function across rows of a 10k-row DataFrame"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 10_000
+WARMUP = 3
+ITERATIONS = 10
+
+a = np.arange(ROWS, dtype=np.float64)
+b = np.arange(ROWS, dtype=np.float64) * 2.0
+df = pd.DataFrame({"a": a, "b": b})
+
+for _ in range(WARMUP):
+    df.apply(lambda row: row["a"] + row["b"], axis=1)
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.apply(lambda row: row["a"] + row["b"], axis=1)
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_apply",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_creation.py

@@ -0,0 +1,27 @@
+"""Benchmark: DataFrame creation from arrays (pandas equivalent)"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 3
+ITERATIONS = 10
+
+nums1 = np.arange(ROWS, dtype=np.float64) * 1.1
+nums2 = np.arange(ROWS, dtype=np.float64) * 2.2
+strs = [f"label_{i % 100}" for i in range(ROWS)]
+
+for _ in range(WARMUP):
+    pd.DataFrame({"a": nums1, "b": nums2, "c": strs})
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    pd.DataFrame({"a": nums1, "b": nums2, "c": strs})
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_creation",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_dropna.py

@@ -0,0 +1,27 @@
+"""Benchmark: dataframe_dropna — drop rows with NaN values from 100k-row DataFrame"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 5
+ITERATIONS = 20
+
+a = np.where(np.arange(ROWS) % 10 == 0, np.nan, np.arange(ROWS) * 1.1)
+b = np.where(np.arange(ROWS) % 7 == 0, np.nan, np.arange(ROWS) * 2.2)
+df = pd.DataFrame({"a": a, "b": b})
+
+for _ in range(WARMUP):
+    df.dropna()
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.dropna()
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_dropna",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_filter.py

@@ -0,0 +1,26 @@
+"""Benchmark: DataFrame filter (boolean mask on 100k-row DataFrame)"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 5
+ITERATIONS = 20
+
+vals = np.arange(ROWS, dtype=np.float64) * 0.1
+df = pd.DataFrame({"value": vals})
+
+for _ in range(WARMUP):
+    df[df["value"] > 5000]
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df[df["value"] > 5000]
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_filter",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_rename.py

@@ -0,0 +1,27 @@
+"""Benchmark: dataframe_rename — rename columns in a 100k-row DataFrame"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 5
+ITERATIONS = 20
+
+a = np.arange(ROWS, dtype=np.float64) * 1.1
+b = np.arange(ROWS, dtype=np.float64) * 2.2
+df = pd.DataFrame({"old_a": a, "old_b": b})
+
+for _ in range(WARMUP):
+    df.rename(columns={"old_a": "new_a", "old_b": "new_b"})
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.rename(columns={"old_a": "new_a", "old_b": "new_b"})
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_rename",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_dataframe_sort.py

@@ -0,0 +1,28 @@
+"""Benchmark: dataframe_sort — sort a 100k-row DataFrame by two columns"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 3
+ITERATIONS = 10
+
+rng = np.random.default_rng(42)
+a = [f"group_{i % 100}" for i in range(ROWS)]
+b = rng.random(ROWS) * 1000
+df = pd.DataFrame({"a": a, "b": b})
+
+for _ in range(WARMUP):
+    df.sort_values(["a", "b"])
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.sort_values(["a", "b"])
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "dataframe_sort",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_describe.py

@@ -0,0 +1,27 @@
+"""Benchmark: describe — summary statistics on a 100k-row DataFrame"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 3
+ITERATIONS = 10
+
+a = np.arange(ROWS, dtype=np.float64) * 1.1
+b = np.sqrt(np.arange(1, ROWS + 1, dtype=np.float64))
+df = pd.DataFrame({"a": a, "b": b})
+
+for _ in range(WARMUP):
+    df.describe()
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.describe()
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "describe",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_ewm_mean.py

@@ -0,0 +1,26 @@
+"""Benchmark: ewm_mean — exponentially weighted mean on 100k-element Series"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 3
+ITERATIONS = 10
+
+data = np.sin(np.arange(ROWS) * 0.05)
+s = pd.Series(data)
+
+for _ in range(WARMUP):
+    s.ewm(span=20).mean()
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    s.ewm(span=20).mean()
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "ewm_mean",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))

benchmarks/pandas/bench_groupby_mean.py

@@ -0,0 +1,27 @@
+"""Benchmark: GroupBy mean on 100k-row DataFrame"""
+import json, time
+import numpy as np
+import pandas as pd
+
+ROWS = 100_000
+WARMUP = 3
+ITERATIONS = 10
+
+keys = [f"group_{i % 100}" for i in range(ROWS)]
+vals = np.arange(ROWS, dtype=np.float64) * 0.1
+df = pd.DataFrame({"key": keys, "value": vals})
+
+for _ in range(WARMUP):
+    df.groupby("key")["value"].mean()
+
+start = time.perf_counter()
+for _ in range(ITERATIONS):
+    df.groupby("key")["value"].mean()
+total = (time.perf_counter() - start) * 1000
+
+print(json.dumps({
+    "function": "groupby_mean",
+    "mean_ms": total / ITERATIONS,
+    "iterations": ITERATIONS,
+    "total_ms": total,
+}))