GreenPolyProducts/cantera
5
0
Fork 0
mirror of https://github.com/Cantera/cantera.git synced 2025-02-25 18:55:29 -06:00
Code Issues Packages Projects Releases Wiki Activity

Refactor CSV and profile tests to better use NumPy

Browse Source
This commit is contained in:
Bryan Weber
2021-04-11 22:54:03 -04:00
committed by Ray Speth
parent 3749acc3bf
commit 87c1a3f70f
1 changed files with 100 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files

176
site_scons/buildutils.py
View File

@@ -350,65 +350,84 @@ def compare_profiles(env, ref_file, sample_file):
After interpolation, each data point must satisfy a combined relative and absolute
error criterion specified by `rtol` and `atol`.
"""
atol = env['test_csv_threshold']
rtol = env['test_csv_tolerance']
xtol = env['test_csv_tolerance']
if not np:
print('WARNING: skipping profile comparison because numpy is not available')
return 0
build_logger.warning("Skipping profile comparison because numpy is not available")
return TestResult.PASS
reference = np.genfromtxt(ref_file, delimiter=',').T
sample = np.genfromtxt(sample_file, delimiter=',').T
assert reference.shape[0] == sample.shape[0]
atol = env["test_csv_threshold"]
rtol = env["test_csv_tolerance"]
xtol = env["test_csv_tolerance"]
# trim header rows if present
if np.isnan(sample[0,0]) and np.isnan(reference[0,0]):
reference = np.genfromtxt(ref_file, delimiter=",").T
sample = np.genfromtxt(sample_file, delimiter=",").T
if reference.shape[0] != sample.shape[0]:
build_logger.error(
"The output array does not have the same number of variabls as the "
"reference array."
)
return TestResult.FAIL
# trim header columns if present
if np.isnan(sample[0, 0]) and np.isnan(reference[0, 0]):
reference = reference[:, 1:]
sample = sample[:, 1:]
assert not np.isnan(reference).any()
assert not np.isnan(sample).any()
if np.isnan(reference).any() or np.isnan(sample).any():
build_logger.error(
"The output array and reference array have different headers "
"or contain non-numeric data."
)
return TestResult.FAIL
nVars = reference.shape[0]
nTimes = reference.shape[1]
n_vars = reference.shape[0]
n_times = reference.shape[1]
bad = []
template = '{0:9.4e} {1: 3d} {2:14.7e} {3:14.7e} {4:9.3e} {5:9.3e} {6:9.3e}'
for i in range(1, nVars):
scale = max(max(abs(reference[i])), reference[i].ptp(),
max(abs(sample[i])), sample[i].ptp())
slope = np.zeros(nTimes)
slope[1:] = np.diff(reference[i]) / np.diff(reference[0]) * reference[0].ptp()
template = "{0:10.4e} {1:5d} {2:14.7e} {3:14.7e} {4:9.3e} {5:9.3e} {6:9.3e}"
header = ["Failed series comparisons:"]
header.append("{:10s} {:5s} {:14s} {:14s} {:9s} {:9s} {:9s}".format(
"coordinate", "comp.", "reference val.", "test value", "abs. err", "rel. err",
"pos. err"
))
header.append(f"{10*'-'} ----- {14*'-'} {14*'-'} {9*'-'} {9*'-'} {9*'-'}")
ref_ptp = reference.ptp(axis=1)
ref_max = np.abs(reference).max(axis=1)
sample_ptp = sample.ptp(axis=1)
sample_max = np.abs(sample).max(axis=1)
scale = np.maximum(
np.maximum(ref_ptp[1:], ref_max[1:]),
np.maximum(sample_ptp[1:], sample_max[1:])
).reshape(n_vars - 1, -1)
ref_diff = np.diff(reference)
slope = ref_diff[1:, :] / ref_diff[0, :] * ref_ptp[0]
slope = np.hstack((np.zeros((n_vars - 1, 1)), slope))
comp = np.zeros((n_vars - 1, n_times))
for i, row in enumerate(sample[1:]):
comp[i, :] = np.interp(reference[0, :], sample[0, :], row)
comp = np.interp(reference[0], sample[0], sample[i])
for j in range(nTimes):
a = reference[i,j]
b = comp[j]
abserr = abs(a-b)
relerr = abs(a-b) / (scale + atol)
# error that can be accounted for by shifting the profile along
# the time / spatial coordinate
xerr = abserr / (abs(slope[j]) + atol)
if abserr > atol and relerr > rtol and xerr > xtol:
bad.append((reference[0][j], i, a, b, abserr, relerr, xerr))
abserr = np.abs(reference[1:] - comp)
relerr = abserr / (scale + atol)
# error that can be accounted for by shifting the profile along
# the time / spatial coordinate
xerr = abserr / (np.abs(slope) + atol)
if np.any(abserr > atol) and np.any(relerr > rtol) and np.any(xerr > xtol):
it = np.nditer((abserr, relerr, xerr), flags=["multi_index"])
for a, r, x in it:
i, j = it.multi_index
bad.append((reference[0, j], i, reference[i,j], comp[i,j], a, r, x))
# Fix line lengths here
footer = []
maxrows = 10
if len(bad) > maxrows:
bad.sort(key=lambda row: -row[5])
footer += ['Plus {0} more points exceeding error thresholds.'.format(len(bad)-maxrows)]
footer += [f"Plus {len(bad) - maxrows} more points exceeding error thresholds."]
bad = bad[:maxrows]
if bad:
header = ['Failed series comparisons:',
'coordinate comp. reference val test value abs. err rel. err pos. err',
'---------- --- -------------- -------------- --------- --------- ---------']
print('\n'.join(header + [template.format(*row) for row in bad] + footer))
return 1
output_logger.error("\n".join(header + [template.format(*row) for row in bad] + footer))
return TestResult.FAIL
else:
return 0
return TestResult.PASS
def compare_csv_files(env, file1: Path, file2: Path):
@@ -428,53 +447,58 @@ def compare_csv_files(env, file1: Path, file2: Path):
files are dissimilar. Lines containing non-numeric data are
automatically ignored.
"""
try:
import numpy as np
except ImportError:
print('WARNING: skipping .csv diff because numpy is not installed')
return 0
if not np:
build_logger.warning("Skipping profile comparison because numpy is not available")
return TestResult.PASS
# decide how many header lines to skip
f = open(file1)
headerRows = 0
goodChars = set('0123456789.+-eE, ')
f = Path(file1).read_text().split("\n")
header_rows = 0
good_chars = set("0123456789.+-eE, ")
for line in f:
if not set(line[:-1]).issubset(goodChars):
headerRows += 1
if not set(line).issubset(good_chars):
header_rows += 1
else:
break
try:
data1 = np.genfromtxt(file1, skip_header=headerRows, delimiter=',')
data2 = np.genfromtxt(file2, skip_header=headerRows, delimiter=',')
data1 = np.genfromtxt(file1, skip_header=header_rows, delimiter=",")
data2 = np.genfromtxt(file2, skip_header=header_rows, delimiter=",")
except (IOError, StopIteration) as e:
print(e)
return 1
build_logger.error(f"Could not read data files: {file1}; {file2}", exc_info=e)
return TestResult.FAIL
threshold = env["test_csv_threshold"]
try:
relerror = (np.abs(data2-data1) /
(np.maximum(np.abs(data2), np.abs(data1)) +
env['test_csv_threshold']))
denom = np.maximum(np.abs(data2), np.abs(data1)) + threshold
relerror = np.abs(data2 - data1) / denom
maxerror = np.nanmax(relerror.flat)
except ValueError as e:
print(e)
return 1
except (ValueError, TypeError) as e:
build_logger.error("Could not compute error.", exc_info=e)
return TestResult.FAIL
tol = env['test_csv_tolerance']
if maxerror > tol: # Threshold based on printing 6 digits in the CSV file
print("Files differ. %i / %i elements above specified tolerance (%f)" %
(np.sum(relerror > tol), relerror.size, tol))
print(' row col reference test rel. error')
print(' ---- ---- -------------- -------------- ----------')
for i,j in itertools.product(*map(range, relerror.shape)):
if relerror[i,j] > tol:
row = i + headerRows + 1
col = j + 1
print(' % 4i % 4i % 14.7e % 14.7e % 10.4e' %
(row, col, data1[i,j], data2[i,j], relerror[i,j]))
return 1
else:
return 0
tol = env["test_csv_tolerance"]
if maxerror < tol: # Threshold based on printing 6 digits in the CSV file
return TestResult.PASS
n_fail = np.sum(relerror > tol)
n_tot = relerror.size
message = [
"Files differ.",
f"{n_fail:d} / {n_tot:d} elements above specified tolerance ({tol:f})",
" row col reference test rel. error",
" ---- ---- -------------- -------------- ----------",
]
it = np.nditer([relerror, data1, data2], flags=["multi_index"])
for rele, ref, test in it:
if rele > tol:
r = it.multi_index[0] + header_rows + 1
c = it.multi_index[1] + 1
message.append(
f" {r:4d} {c:4d} {ref:14.7e} {test:14.7e} {rele:10.4e}"
)
build_logger.error("\n".join(message))
return TestResult.FAIL
def regression_test_message(target, source, env):