datarust 0.6.5

Scikit-learn-style preprocessing and classical ML in Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149

#!/usr/bin/env python3
"""Benchmark mirror of examples/bench_compare_rust.rs.

Generates deterministic synthetic data with the *same* PRNG layout as the Rust
runner (xorshift64, seed 42) so both sides transform identical inputs, then
reports median `fit_transform` time in milliseconds.

Usage: python3 benches/compare_sklearn.py [reps]
"""

import sys
import time

import numpy as np
from sklearn.decomposition import PCA
from sklearn.linear_model import LinearRegression as SkLinearRegression
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import (
    MinMaxScaler,
    OneHotEncoder,
    RobustScaler,
    StandardScaler,
)
from sklearn.compose import ColumnTransformer


# --- xorshift64 PRNG identical to the Rust runner -----------------------------

class XorShift64:
    def __init__(self, seed: int):
        self.state = 0x9E3779B97F4A7C15 if seed == 0 else seed

    def next_u64(self) -> int:
        s = self.state
        s ^= (s << 13) & 0xFFFFFFFFFFFFFFFF
        s ^= s >> 7
        s ^= (s << 17) & 0xFFFFFFFFFFFFFFFF
        self.state = s
        return s

    def next_unit(self) -> float:
        return (self.next_u64() >> 11) / float(1 << 53)

    def next_range(self, lo: float, hi: float) -> float:
        return lo + (hi - lo) * self.next_unit()


def make_matrix(rows: int, cols: int, seed: int = 42) -> np.ndarray:
    rng = XorShift64(seed)
    # Generate in the same iteration order as Rust: row-major, col inner.
    flat = [rng.next_range(-100.0, 100.0) for _ in range(rows * cols)]
    return np.asarray(flat, dtype=np.float64).reshape(rows, cols)


def make_str_array(rows: int, cols: int, seed: int = 42, cardinality: int = 20):
    rng = XorShift64(seed)
    # List-of-lists of strings, same iteration order as Rust.
    data = [
        [f"cat_{int(rng.next_unit() * cardinality)}" for _ in range(cols)]
        for _ in range(rows)
    ]
    # object array so sklearn treats each cell as a string.
    return np.asarray(data, dtype=object)


def median(samples):
    s = sorted(samples)
    n = len(s)
    return s[n // 2] if n % 2 else (s[n // 2 - 1] + s[n // 2]) / 2.0


def measure(reps: int, workload):
    workload()  # warmup
    samples = []
    for _ in range(reps):
        t0 = time.perf_counter()
        workload()
        dt = (time.perf_counter() - t0) * 1000.0
        samples.append(dt)
    return median(samples)


def main():
    reps = int(sys.argv[1]) if len(sys.argv) > 1 else 15
    sizes = [(1_000, 10), (10_000, 100), (50_000, 200)]
    cat_sizes = [(1_000, 5), (10_000, 10), (50_000, 20)]

    print("workload,rows,cols,sklearn_ms")
    for rows, cols in sizes:
        x = make_matrix(rows, cols, 42)

        ms = measure(reps, lambda: StandardScaler().fit_transform(x))
        print(f"standard_scaler,{rows},{cols},{ms:.4f}")

        ms = measure(reps, lambda: MinMaxScaler().fit_transform(x))
        print(f"minmax_scaler,{rows},{cols},{ms:.4f}")

        ms = measure(reps, lambda: RobustScaler().fit_transform(x))
        print(f"robust_scaler,{rows},{cols},{ms:.4f}")

        k = min(10, max(1, cols // 2))
        ms = measure(reps, lambda: PCA(n_components=k).fit_transform(x))
        print(f"pca,{rows},{cols},{ms:.4f}")

        # LinearRegression — same deterministic y as the Rust runner.
        y = x @ np.arange(1, cols + 1, dtype=np.float64)
        ms = measure(
            reps,
            lambda: SkLinearRegression().fit(x, y).predict(x),
        )
        print(f"linear_regression,{rows},{cols},{ms:.4f}")

        ms = measure(
            reps,
            lambda: Pipeline(
                [
                    ("s1", StandardScaler()),
                    ("s2", MinMaxScaler()),
                    ("s3", RobustScaler()),
                ]
            ).fit_transform(x),
        )
        print(f"pipeline_3scalers,{rows},{cols},{ms:.4f}")

    for rows, cols in cat_sizes:
        x_str = make_str_array(rows, cols, 42, 20)
        ms = measure(reps, lambda: OneHotEncoder(sparse_output=False).fit_transform(x_str))
        print(f"onehot_encoder,{rows},{cols},{ms:.4f}")

        # ColumnTransformer: numeric scaled + categorical one-hot.
        numeric = make_matrix(rows, cols, 7)
        categorical = make_str_array(rows, cols, 11, 15)
        x_all = np.hstack([numeric, categorical.astype(str)])
        num_idx = list(range(cols))
        cat_idx = list(range(cols, 2 * cols))
        ms = measure(
            reps,
            lambda: ColumnTransformer(
                transformers=[
                    ("num", StandardScaler(), num_idx),
                    ("cat", OneHotEncoder(sparse_output=False), cat_idx),
                ]
            ).fit_transform(x_all),
        )
        print(f"column_transformer,{rows},{cols},{ms:.4f}")


if __name__ == "__main__":
    main()