DAMASK_EICMD/python/tests/test_util.py

47 lines
1.6 KiB
Python
Raw Normal View History

import pytest
import numpy as np
2020-09-28 11:10:43 +05:30
from scipy import stats
2020-06-21 02:21:00 +05:30
from damask import util
2020-06-21 02:21:00 +05:30
class TestUtil:
def test_execute_direct(self):
out,err = util.execute('echo test')
assert out=='test\n' and err==''
2020-06-21 02:21:00 +05:30
def test_execute_env(self):
out,err = util.execute('sh -c "echo $test_for_execute"',env={'test_for_execute':'test'})
assert out=='test\n' and err==''
def test_croak(self):
util.croak('Burp!')
@pytest.mark.parametrize('input,output',
[
([0,-2],[0,-1]),
([-0.5,0.5],[-1,1]),
([1./2.,1./3.],[3,2]),
([2./3.,1./2.,1./3.],[4,3,2]),
])
def test_scale2coprime(self,input,output):
assert np.allclose(util.scale_to_coprime(np.array(input)),
np.array(output).astype(int))
def test_lackofprecision(self):
with pytest.raises(ValueError):
util.scale_to_coprime(np.array([1/333.333,1,1]))
2020-09-28 11:10:43 +05:30
@pytest.mark.parametrize('rv',[stats.rayleigh(),stats.weibull_min(1.2),stats.halfnorm(),stats.pareto(2.62)])
def test_hybridIA(self,rv):
bins = np.linspace(0,10,100000)
centers = (bins[1:]+bins[:-1])/2
N_samples = bins.shape[0]-1000
dist = rv.pdf(centers)
selected = util.hybrid_IA(dist,N_samples)
dist_sampled = np.histogram(centers[selected],bins)[0]/N_samples*np.sum(dist)
assert np.sqrt(((dist - dist_sampled) ** 2).mean()) < .025 and selected.shape[0]==N_samples