lattice/pyerrors
1
0
Fork 0
mirror of https://github.com/fjosw/pyerrors.git synced 2025-05-14 19:43:41 +02:00
Code Issues Projects Releases Packages Wiki Activity

[fix] Handle missing replia (#232)

Browse source 
* [fix] First version of a fix to cope with missing replica.

* [fix] added test for missing replica

* [fix] refactored fix for missing replica, modified tests

* [fix] refinement of tests
This commit is contained in:
s-kuberski 2024-04-25 20:45:53 +02:00 • committed by GitHub
parent 43bd99b6c7
commit db612597d2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 124 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

93
tests/obs_test.py
View file

@ -5,6 +5,7 @@ import copy
import matplotlib.pyplot as plt
import pyerrors as pe
import pytest
import pyerrors.linalg
from hypothesis import given, strategies as st
np.random.seed(0)
@ -1338,9 +1339,101 @@ def test_vec_gm():
pe.gm(cc, S=4.12)
assert np.all(np.vectorize(lambda x: x.S["qq"])(cc.content) == 4.12)
def test_complex_addition():
o = pe.pseudo_Obs(34.12, 1e-4, "testens")
r = o + 2j
assert r.real == o
r = r * 1j
assert r.imag == o
def test_missing_replica():
N1 = 3000
N2 = 2000
O1 = np.random.normal(1.0, .1, N1 + N2)
O2 = .5 * O1[:N1]
w1 = N1 / (N1 + N2)
w2 = N2 / (N1 + N2)
m12 = np.mean(O1[N1:])
m2 = np.mean(O2)
d12 = np.std(O1[N1:]) / np.sqrt(N2) # error of <O1> from second rep
d2 = np.std(O2) / np.sqrt(N1) # error of <O2> from first rep
dval = np.sqrt((w2 * d12 / m2)**2 + (w2 * m12 * d2 / m2**2)**2) # complete error of <O1>/<O2>
# pyerrors version that should give the same result
O1dobs = pe.Obs([O1[:N1], O1[N1:]], names=['E|1', 'E|2'])
O2dobs = pe.Obs([O2], names=['E|1'])
O1O2 = O1dobs / O2dobs
O1O2.gm(S=0)
# explicit construction with different ensembles
O1a = pe.Obs([O1[:N1]], names=['E|1'])
O1b = pe.Obs([O1[N1:]], names=['F|2'])
O1O2b = (w1 * O1a + w2 * O1b) / O2dobs
O1O2b.gm(S=0)
# pyerrors version without replica (missing configs)
O1c = pe.Obs([O1], names=['E|1'])
O1O2c = O1c / O2dobs
O1O2c.gm(S=0)
for o in [O1O2, O1O2b, O1O2c]:
assert(np.isclose(dval, o.dvalue, atol=0, rtol=5e-2))
o = O1O2 * O2dobs - O1dobs
o.gm()
assert(o.is_zero())
o = O1dobs / O1O2 - O2dobs
o.gm()
assert(o.is_zero())
# bring more randomness and complexity into the game
Nl = [int(np.random.uniform(low=500, high=5000)) for i in range(4)]
wl = np.array(Nl) / sum(Nl)
O1 = np.random.normal(1.0, .1, sum(Nl))
# pyerrors replica version
datl = [O1[:Nl[0]], O1[Nl[0]:sum(Nl[:2])], O1[sum(Nl[:2]):sum(Nl[:3])], O1[sum(Nl[:3]):sum(Nl[:4])]]
O1dobs = pe.Obs(datl, names=['E|%d' % (d) for d in range(len(Nl))])
O2dobs = .5 * pe.Obs([datl[0]], names=['E|0'])
O3dobs = 2. / pe.Obs([datl[1]], names=['E|1'])
O1O2 = O1dobs / O2dobs
O1O2.gm(S=0)
O1O2O3 = O1O2 * np.sinh(O3dobs)
O1O2O3.gm(S=0)
# explicit construction with different ensembles
charl = ['E', 'F', 'G', 'H']
Ol = [pe.Obs([datl[i]], names=['%s|%d' % (charl[i], i)]) for i in range(len(Nl))]
O1O2b = sum(np.array(Ol) * wl) / O2dobs
O1O2b.gm(S=0)
i = 1
O3dobsb = 2. / pe.Obs([datl[i]], names=['%s|%d' % (charl[i], i)])
O1O2O3b = O1O2b * np.sinh(O3dobsb)
O1O2O3b.gm(S=0)
for op in [[O1O2, O1O2b], [O1O2O3, O1O2O3b]]:
assert np.isclose(op[0].value, op[1].value)
assert np.isclose(op[0].dvalue, op[1].dvalue, atol=0, rtol=5e-2)
# perform the same test using the array_mode of derived_observable
O1O2 = pyerrors.linalg.matmul(np.diag(np.diag(np.reshape(4 * [O1dobs], (2, 2)))), np.diag(np.diag(np.reshape(4 * [1. / O2dobs], (2, 2)))))
O1O2O3 = pyerrors.linalg.matmul(O1O2, np.diag(np.diag(np.sinh(np.reshape(4 * [O3dobs], (2, 2))))))
O1O2 = O1O2[0][0]
O1O2.gm(S=0)
O1O2O3 = O1O2O3[0][0]
O1O2O3.gm(S=0)
O1O2b = pyerrors.linalg.matmul(np.diag(np.diag(np.reshape(4 * [sum(np.array(Ol) * wl)], (2, 2)))), np.diag(np.diag(np.reshape(4 * [1. / O2dobs], (2, 2)))))
O1O2O3b = pyerrors.linalg.matmul(O1O2b, np.diag(np.diag(np.sinh(np.reshape(4 * [O3dobsb], (2, 2))))))
O1O2b = O1O2b[0][0]
O1O2b.gm(S=0)
O1O2O3b = O1O2O3b[0][0]
O1O2O3b.gm(S=0)
for op in [[O1O2, O1O2b], [O1O2O3, O1O2O3b]]:
assert np.isclose(op[1].value, op[0].value)
assert np.isclose(op[1].dvalue, op[0].dvalue, atol=0, rtol=5e-2)