inferust 0.1.21

Statistical modeling for Rust — OLS/WLS regression, GLM, survival analysis, ARIMA/VAR, nonparametric tests, and more. A statsmodels-style library.
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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167

#!/usr/bin/env python3
"""
Full benchmark suite: statsmodels / scipy / lifelines equivalents for every
estimator covered by bench_all.rs. Outputs key=value lines.
"""
import time, warnings
import numpy as np
import pandas as pd
import statsmodels.api as sm
from statsmodels.regression.quantile_regression import QuantReg
from statsmodels.regression.rolling import RollingOLS
from statsmodels.regression.recursive_ls import RecursiveLS
from lifelines import CoxPHFitter
from scipy import stats as spstats
warnings.filterwarnings("ignore")

RNG = np.random.default_rng(42)
N = 10_000

def ms(s): return s * 1000.0

def bench(label, fn, rows=N, repeats=20, warmups=3):
    for _ in range(warmups):
        try: fn()
        except: pass
    times = []
    for _ in range(repeats):
        t0 = time.perf_counter()
        try: fn()
        except: pass
        times.append(time.perf_counter() - t0)
    times.sort()
    print(f"engine=python-statsmodels estimator={label} rows={rows} repeats={repeats} "
          f"min_ms={ms(times[0]):.3f} median_ms={ms(times[len(times)//2]):.3f} "
          f"mean_ms={ms(sum(times)/len(times)):.3f}", flush=True)

# ── OLS ──────────────────────────────────────────────────────────────────────
X = RNG.standard_normal((N,8)); y = X @ np.arange(1,9,dtype=float) + RNG.standard_normal(N)
Xc = sm.add_constant(X)
bench("ols", lambda: sm.OLS(y, Xc).fit())

# ── WLS ──────────────────────────────────────────────────────────────────────
X3 = RNG.standard_normal((N,3)); y3 = X3 @ [1.,2.,3.] + RNG.standard_normal(N)
w = 1/(1+np.arange(N)*0.0001); X3c = sm.add_constant(X3)
bench("wls", lambda: sm.WLS(y3, X3c, weights=w).fit())

# ── Ridge (via statsmodels fit_regularized) ──────────────────────────────────
X4 = RNG.standard_normal((N,4)); y4 = X4 @ [1.,2.,3.,4.] + RNG.standard_normal(N)
X4c = sm.add_constant(X4)
bench("ridge", lambda: sm.OLS(y4, X4c).fit_regularized(method='elastic_net', alpha=0.1, L1_wt=0.0))

# ── Lasso ────────────────────────────────────────────────────────────────────
bench("lasso", lambda: sm.OLS(y4, X4c).fit_regularized(method='elastic_net', alpha=0.01, L1_wt=1.0))

# ── ElasticNet ───────────────────────────────────────────────────────────────
bench("elastic_net", lambda: sm.OLS(y4, X4c).fit_regularized(method='elastic_net', alpha=0.01, L1_wt=0.5))

# ── QuantileRegression ───────────────────────────────────────────────────────
bench("quantile_reg", lambda: QuantReg(y3, X3c).fit(q=0.5))

# ── RollingOLS ───────────────────────────────────────────────────────────────
X2 = RNG.standard_normal((N,2)); y2 = X2 @ [1.,2.] + RNG.standard_normal(N)
X2c = sm.add_constant(X2)
bench("rolling_ols", lambda: RollingOLS(y2, X2c, window=200).fit(), repeats=10)

# ── RecursiveOLS ─────────────────────────────────────────────────────────────
bench("recursive_ols", lambda: RecursiveLS(y2, X2c).fit(), repeats=10)

# ── FGLS (Cochrane-Orcutt via GLSAR) ─────────────────────────────────────────
bench("fgls", lambda: sm.GLSAR(y3, X3c, rho=1).iterative_fit(maxiter=20), repeats=10)

# ── RobustLM (IRLS Huber via RLM) ────────────────────────────────────────────
bench("robust_lm", lambda: sm.RLM(y3, X3c, M=sm.robust.norms.HuberT()).fit(), repeats=10)

# ── Logistic ─────────────────────────────────────────────────────────────────
Xl = RNG.standard_normal((N,4))
pl = 1/(1+np.exp(-(Xl @ [0.5,-1,1.5,-0.5])))
yl = RNG.binomial(1, pl).astype(float)
Xlc = sm.add_constant(Xl)
bench("logistic", lambda: sm.GLM(yl, Xlc, family=sm.families.Binomial()).fit())

# ── Poisson ───────────────────────────────────────────────────────────────────
Xp = RNG.standard_normal((N,4))
yp = RNG.poisson(np.exp(Xp @ [0.3,0.5,-0.2,0.4] + 1.0)).astype(float)
Xpc = sm.add_constant(Xp)
bench("poisson", lambda: sm.GLM(yp, Xpc, family=sm.families.Poisson()).fit())

# ── Gamma GLM ─────────────────────────────────────────────────────────────────
Xg = RNG.standard_normal((N,3))
yg = np.exp(Xg @ [0.3,0.5,-0.2] + 1.0) * (1 + 0.1*np.abs(RNG.standard_normal(N)))
Xgc = sm.add_constant(Xg)
bench("gamma_glm", lambda: sm.GLM(yg, Xgc, family=sm.families.Gamma(sm.families.links.Log())).fit())

# ── Probit ────────────────────────────────────────────────────────────────────
Xpr = RNG.standard_normal((N,3))
ypr = (Xpr @ [0.8,-0.5,1.0] + RNG.standard_normal(N) > 0).astype(float)
Xprc = sm.add_constant(Xpr)
bench("probit", lambda: sm.Probit(ypr, Xprc).fit(disp=False))

# ── NegativeBinomial ─────────────────────────────────────────────────────────
ynb = RNG.poisson(np.exp(Xg @ [0.3,0.5,-0.2] + 1.0)).astype(float)
bench("neg_binomial", lambda: sm.NegativeBinomial(ynb, Xgc).fit(disp=False), repeats=10, warmups=1)

# ── MultinomialLogit ─────────────────────────────────────────────────────────
Xm = RNG.standard_normal((N,3))
eta1 = Xm[:,0]*0.5 - Xm[:,1]*0.3
eta2 = -Xm[:,0]*0.2 + Xm[:,2]*0.8
denom = 1 + np.exp(eta1) + np.exp(eta2)
p = np.column_stack([1/denom, np.exp(eta1)/denom, np.exp(eta2)/denom])
cum = p.cumsum(axis=1); u = RNG.random(N)[:,None]
ym = (cum < u).sum(axis=1)
Xmc = sm.add_constant(Xm)
bench("mnlogit", lambda: sm.MNLogit(ym, Xmc).fit(disp=False), repeats=10, warmups=1)

# ── OrderedLogit ─────────────────────────────────────────────────────────────
from statsmodels.miscmodels.ordinal_model import OrderedModel
Xo = RNG.standard_normal((N,2))
eta_o = Xo[:,0]*0.8 - Xo[:,1]*0.4 + 0.3*RNG.standard_normal(N)
yo = pd.Categorical(np.where(eta_o < -0.5, 0, np.where(eta_o < 0.5, 1, 2)))
bench("ordered_logit", lambda: OrderedModel(yo, Xo, distr='logit').fit(method='bfgs', disp=False), repeats=5, warmups=1)

# ── ZeroInflatedPoisson ──────────────────────────────────────────────────────
from statsmodels.discrete.count_model import ZeroInflatedPoisson as SM_ZIP
Xz = RNG.standard_normal((N,2))
mu_z = np.exp(Xz @ [0.5,-0.3] + 1.0)
mask = RNG.random(N) < 0.2
yz = np.where(mask, 0, RNG.poisson(mu_z)).astype(float)
Xzc = sm.add_constant(Xz)
bench("zip", lambda: SM_ZIP(yz, Xzc).fit(disp=False), repeats=5, warmups=1)

# ── GEE ──────────────────────────────────────────────────────────────────────
Xe = RNG.standard_normal((N,3))
ye = RNG.poisson(np.exp(Xe @ [0.5,-0.3,0.4] + 0.5)).astype(float)
clusters_e = np.repeat(np.arange(N//10), 10)
Xec = sm.add_constant(Xe)
bench("gee", lambda: sm.GEE(ye, Xec, groups=clusters_e, family=sm.families.Poisson(),
      cov_struct=sm.cov_struct.Exchangeable()).fit(), repeats=5, warmups=1)

# ── MixedLM ──────────────────────────────────────────────────────────────────
Xmx = RNG.standard_normal((N,2))
ymx = Xmx @ [1.,2.] + RNG.standard_normal(N)
groups_mx = np.repeat(np.arange(N//20), 20)
Xmxc = sm.add_constant(Xmx)
bench("mixed_lm", lambda: sm.MixedLM(ymx, Xmxc, groups=groups_mx).fit(), repeats=5, warmups=1)

# ── KaplanMeier (lifelines) ───────────────────────────────────────────────────
from lifelines import KaplanMeierFitter
t_km = RNG.exponential(1, N); e_km = RNG.binomial(1, 0.8, N)
kmf = KaplanMeierFitter()
bench("kaplan_meier", lambda: kmf.fit(t_km, e_km))

# ── CoxPH ────────────────────────────────────────────────────────────────────
Xcox = RNG.standard_normal((N,3))
lam = np.exp(Xcox @ [0.5,-0.3,0.8])
t_cox = RNG.exponential(1/lam); e_cox = RNG.binomial(1, 0.8, N)
df_cox = pd.DataFrame(Xcox, columns=["x1","x2","x3"])
df_cox["T"] = t_cox; df_cox["E"] = e_cox
cph = CoxPHFitter()
bench("cox_ph", lambda: cph.fit(df_cox, "T", "E"), repeats=5, warmups=1)

# ── t-test (two-sample) ──────────────────────────────────────────────────────
a_t = RNG.standard_normal(N); b_t = RNG.standard_normal(N) + 0.2
bench("ttest_2samp", lambda: spstats.ttest_ind(a_t, b_t))

# ── Mann-Whitney U ───────────────────────────────────────────────────────────
a_mw = RNG.standard_normal(1000); b_mw = RNG.standard_normal(1000) + 0.2
bench("mann_whitney", lambda: spstats.mannwhitneyu(a_mw, b_mw), rows=1000)