neopdf_pyapi 0.3.3

Python bindings to NeoPDF Rust 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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196

"""Tests for neopdf.uncertainty, benchmarked against LHAPDF."""

import numpy as np
import pytest
import lhapdf

from neopdf.pdf import PDF as NeoPDF
from neopdf.uncertainty import CL_1_SIGMA, CL_2_SIGMA, uncertainty, Uncertainty


def lha_values(pdfname: str, pid: int, x: float, q: float) -> np.ndarray:
    """Return an array of xfxQ values (all members) from LHAPDF."""
    pdfs = lhapdf.mkPDFs(pdfname)
    return np.array([p.xfxQ(pid, x, q) for p in pdfs])


def neo_values(pdfname: str, pid: int, x: float, q: float) -> np.ndarray:
    """Return an array of xfxQ2 values (all members) from NeoPDF."""
    pdfs = NeoPDF.mkPDFs(pdfname)
    return np.array([p.xfxQ2(pid, x, q * q) for p in pdfs])


class TestUncertaintyClass:
    def test_attributes(self):
        values = np.array([1.0, 0.9, 1.1, 0.95, 1.05])
        unc = uncertainty(values, "replicas", cl=CL_1_SIGMA)
        assert hasattr(unc, "central")
        assert hasattr(unc, "errminus")
        assert hasattr(unc, "errplus")
        assert isinstance(unc, Uncertainty)

    def test_central_replicas_is_mean(self):
        values = np.array([5.0, 4.0, 6.0, 4.5, 5.5])
        unc = uncertainty(values, "replicas", cl=CL_1_SIGMA)
        assert unc.central == np.mean(values[1:])

    def test_errsymm(self):
        values = np.array([1.0, 0.8, 1.2, 0.9, 1.1])
        unc = uncertainty(values, "hessian", cl=CL_1_SIGMA)
        assert np.isclose(unc.errsymm(), (unc.errminus + unc.errplus) / 2.0)

    def test_repr(self):
        values = np.array([1.0, 0.9, 1.1])
        unc = uncertainty(values, "replicas", cl=CL_1_SIGMA)
        assert "Uncertainty" in repr(unc)
        assert "central" in repr(unc)

    def test_raises_on_empty(self):
        with pytest.raises(Exception):
            uncertainty(np.array([]), "replicas", cl=CL_1_SIGMA)


class TestUncertaintyAlgorithms:
    def test_replicas_symmetric(self):
        rng = np.random.default_rng(42)
        replicas = rng.normal(loc=0.0, scale=1.0, size=100)
        values = np.concatenate([[0.0], replicas])
        unc = uncertainty(values, "replicas", cl=CL_1_SIGMA)
        expected_std = np.std(replicas, ddof=1)
        assert np.isclose(unc.errminus, expected_std, rtol=1e-10)
        assert np.isclose(unc.errplus, expected_std, rtol=1e-10)

    def test_replicas_alternative_asymmetric(self):
        rng = np.random.default_rng(0)
        replicas = rng.exponential(scale=1.0, size=1000)
        values = np.concatenate([[np.mean(replicas)], replicas])
        unc_sym = uncertainty(values, "replicas", cl=68.27)
        unc_alt = uncertainty(values, "replicas", cl=68.27, alternative=True)
        assert not np.isclose(unc_alt.errminus, unc_alt.errplus, rtol=1e-3)
        assert np.isclose(unc_sym.errminus, unc_sym.errplus, rtol=1e-10)

    def test_hessian_symmetric_pairs(self):
        values = np.array([1.0, 1.1, 0.9, 1.2, 0.8])
        unc = uncertainty(values, "hessian", cl=CL_1_SIGMA)
        expected = np.sqrt(0.1**2 + 0.2**2)
        assert np.isclose(unc.errminus, expected, rtol=1e-10)
        assert np.isclose(unc.errplus, expected, rtol=1e-10)

    def test_symmhessian_same_as_hessian(self):
        values = np.array([1.0, 1.1, 0.9, 1.3, 0.7])
        unc_h = uncertainty(values, "hessian", cl=CL_1_SIGMA)
        unc_s = uncertainty(values, "symmhessian", cl=CL_1_SIGMA)
        assert np.isclose(unc_h.errminus, unc_s.errminus)
        assert np.isclose(unc_h.errplus, unc_s.errplus)

    def test_asymhessian(self):
        values = np.array([1.0, 1.2, 0.8])
        unc = uncertainty(values, "asymhessian", cl=CL_1_SIGMA)
        assert np.isclose(unc.errplus, 0.2, rtol=1e-10)
        assert np.isclose(unc.errminus, 0.2, rtol=1e-10)

    def test_cl_rescaling_hessian(self):
        values = np.array([1.0, 1.1, 0.9, 1.2, 0.8])
        unc_1s = uncertainty(values, "hessian", cl=68.27)
        unc_2s = uncertainty(values, "hessian", cl=95.45)
        assert np.isclose(unc_2s.errminus / unc_1s.errminus, 2.0, rtol=1e-2)

    def test_cl_rescaling_replicas(self):
        rng = np.random.default_rng(7)
        replicas = rng.normal(0.0, 1.0, 500)
        values = np.concatenate([[0.0], replicas])
        unc_1s = uncertainty(values, "replicas", cl=68.27)
        unc_2s = uncertainty(values, "replicas", cl=95.45)
        assert np.isclose(unc_2s.errminus / unc_1s.errminus, 2.0, rtol=1e-2)

    def test_error_conf_level_ct18_convention(self):
        values = np.array([1.0, 1.2, 0.8, 1.3, 0.7])
        unc_68 = uncertainty(
            values,
            "hessian",
            error_conf_level=CL_1_SIGMA,
            cl=CL_1_SIGMA,
        )
        unc_90 = uncertainty(
            values,
            "hessian",
            error_conf_level=90.0,
            cl=CL_1_SIGMA,
        )
        assert unc_90.errminus < unc_68.errminus
        assert np.isclose(
            unc_90.errminus / unc_68.errminus,
            1.0 / (1.6449 / 1.0),
            rtol=1e-2,
        )


@pytest.mark.parametrize(
    "pdfname",
    ["NNPDF40_nnlo_as_01180", "CT18NNLO_as_0118", "MSHT20qed_an3lo"],
)
@pytest.mark.parametrize(
    "pid,x,q",
    [(21, 0.1, 100.0), (2, 0.01, 10.0), (-3, 0.3, 1000.0)],
)
class TestAgainstLhapdf:
    def _setup(self, pdfname, pid, x, q):
        pdfset = lhapdf.getPDFSet(pdfname)
        values = lha_values(pdfname, pid, x, q)

        if len(values) <= 1:
            pytest.skip(f"{pdfname} has only {len(values)} member(s) installed")

        pdf0 = NeoPDF(pdfname, 0)
        error_type = pdf0.metadata().error_type()
        ecl = float(pdfset.errorConfLevel)  # -1.0 when not set in the info file
        return pdfset, values, error_type, ecl

    def test_central(self, pdfname, pid, x, q):
        pdfset, values, error_type, ecl = self._setup(pdfname, pid, x, q)
        lha = pdfset.uncertainty(values, CL_1_SIGMA, False)
        neo = uncertainty(values, error_type, error_conf_level=ecl, cl=CL_1_SIGMA)
        np.testing.assert_allclose(neo.central, lha.central, rtol=1e-10)

    def test_errminus(self, pdfname, pid, x, q):
        pdfset, values, error_type, ecl = self._setup(pdfname, pid, x, q)
        lha = pdfset.uncertainty(values, CL_1_SIGMA, False)
        neo = uncertainty(values, error_type, error_conf_level=ecl, cl=CL_1_SIGMA)
        # TODO: To be investigated further
        np.testing.assert_allclose(neo.errminus, lha.errminus, rtol=1e-2)

    def test_errplus(self, pdfname, pid, x, q):
        pdfset, values, error_type, ecl = self._setup(pdfname, pid, x, q)
        lha = pdfset.uncertainty(values, CL_1_SIGMA, False)
        neo = uncertainty(values, error_type, error_conf_level=ecl, cl=CL_1_SIGMA)
        np.testing.assert_allclose(neo.errplus, lha.errplus, rtol=1e-3)

    def test_alternative_prescription(self, pdfname, pid, x, q):
        if pdfname != "NNPDF40_nnlo_as_01180":
            pytest.skip("alternative prescription only applies to replica sets")
        pdfset, values, error_type, ecl = self._setup(pdfname, pid, x, q)
        neo = uncertainty(
            values,
            error_type,
            error_conf_level=ecl,
            cl=CL_1_SIGMA,
            alternative=True,
        )
        neo_std = uncertainty(
            values,
            error_type,
            error_conf_level=ecl,
            cl=CL_1_SIGMA,
        )
        assert neo.errminus != neo.errplus, "expected asymmetric interval"
        assert neo.errminus > 0
        assert neo.errplus > 0
        assert neo.errminus < 3 * neo_std.errminus
        assert neo.errplus < 3 * neo_std.errplus

    def test_higher_cl(self, pdfname, pid, x, q):
        pdfset, values, error_type, ecl = self._setup(pdfname, pid, x, q)
        neo_1s = uncertainty(values, error_type, error_conf_level=ecl, cl=CL_1_SIGMA)
        neo_2s = uncertainty(values, error_type, error_conf_level=ecl, cl=CL_2_SIGMA)
        assert neo_2s.errminus >= neo_1s.errminus
        assert neo_2s.errplus >= neo_1s.errplus