Files
galvani/tests/test_BioLogic.py

380 lines
13 KiB
Python

# -*- coding: utf-8 -*-
# SPDX-FileCopyrightText: 2013-2020 Christopher Kerr <chris.kerr@mykolab.ch>
#
# SPDX-License-Identifier: GPL-3.0-or-later
import os.path
import re
from datetime import date, datetime
import numpy as np
from numpy.testing import assert_array_almost_equal, assert_array_equal, assert_allclose
import pytest
from galvani import BioLogic, MPTfile, MPRfile
from galvani.BioLogic import MPTfileCSV # not exported
def test_open_MPT(testdata_dir):
mpt1, comments = MPTfile(os.path.join(testdata_dir, "bio_logic1.mpt"))
assert comments == []
assert mpt1.dtype.names == (
"mode",
"ox/red",
"error",
"control changes",
"Ns changes",
"counter inc.",
"time/s",
"control/V/mA",
"Ewe/V",
"dQ/mA.h",
"P/W",
"I/mA",
"(Q-Qo)/mA.h",
"x",
)
def test_open_MPT_fails_for_bad_file(testdata_dir):
with pytest.raises(ValueError, match="Bad first line"):
MPTfile(os.path.join(testdata_dir, "bio_logic1.mpr"))
def test_open_MPT_csv(testdata_dir):
mpt1, comments = MPTfileCSV(os.path.join(testdata_dir, "bio_logic1.mpt"))
assert comments == []
assert mpt1.fieldnames == [
"mode",
"ox/red",
"error",
"control changes",
"Ns changes",
"counter inc.",
"time/s",
"control/V/mA",
"Ewe/V",
"dq/mA.h",
"P/W",
"<I>/mA",
"(Q-Qo)/mA.h",
"x",
]
def test_open_MPT_csv_fails_for_bad_file(testdata_dir):
with pytest.raises((ValueError, UnicodeDecodeError)):
MPTfileCSV(os.path.join(testdata_dir, "bio_logic1.mpr"))
def test_colID_map_uniqueness():
"""Check some uniqueness properties of the VMPdata_colID_xyz maps."""
field_colIDs = set(BioLogic.VMPdata_colID_dtype_map.keys())
flag_colIDs = set(BioLogic.VMPdata_colID_flag_map.keys())
field_names = [v[0] for v in BioLogic.VMPdata_colID_dtype_map.values()]
flag_names = [v[0] for v in BioLogic.VMPdata_colID_flag_map.values()]
assert not field_colIDs.intersection(flag_colIDs)
# 'I/mA' and 'dQ/mA.h' are duplicated
# assert len(set(field_names)) == len(field_names)
assert len(set(flag_names)) == len(flag_names)
assert not set(field_names).intersection(flag_names)
@pytest.mark.parametrize(
"colIDs, expected",
[
([1, 2, 3], [("flags", "u1")]),
([4, 6], [("time/s", "<f8"), ("Ewe/V", "<f4")]),
([1, 4, 21], [("flags", "u1"), ("time/s", "<f8")]),
([4, 6, 4], [("time/s", "<f8"), ("Ewe/V", "<f4"), ("time/s 2", "<f8")]),
([4, 9999], NotImplementedError),
],
)
def test_colID_to_dtype(colIDs, expected):
"""Test converting column ID to numpy dtype."""
if isinstance(expected, type) and issubclass(expected, Exception):
with pytest.raises(expected):
BioLogic.VMPdata_dtype_from_colIDs(colIDs)
return
expected_dtype = np.dtype(expected)
dtype, flags_dict = BioLogic.VMPdata_dtype_from_colIDs(colIDs)
assert np.dtype(dtype) == expected_dtype
@pytest.mark.parametrize(
"data, expected",
[
("02/23/17", date(2017, 2, 23)),
("10-03-05", date(2005, 10, 3)),
("11.12.20", date(2020, 11, 12)),
(b"01/02/03", date(2003, 1, 2)),
("13.08.07", ValueError),
("03-04/05", ValueError),
],
)
def test_parse_BioLogic_date(data, expected):
"""Test the parse_BioLogic_date function."""
if isinstance(expected, type) and issubclass(expected, Exception):
with pytest.raises(expected):
BioLogic.parse_BioLogic_date(data)
return
result = BioLogic.parse_BioLogic_date(data)
assert result == expected
@pytest.mark.parametrize(
"filename, startdate, enddate",
[
("bio_logic1.mpr", "2011-10-29", "2011-10-31"),
("bio_logic2.mpr", "2012-09-27", "2012-09-27"),
("bio_logic3.mpr", "2013-03-27", "2013-03-27"),
("bio_logic4.mpr", "2011-11-01", "2011-11-02"),
("bio_logic5.mpr", "2013-01-28", "2013-01-28"),
# bio_logic6.mpr has no end date because it does not have a VMP LOG module
("bio_logic6.mpr", "2012-09-11", None),
# C019P-0ppb-A_C01.mpr stores the date in a different format
("C019P-0ppb-A_C01.mpr", "2019-03-14", "2019-03-14"),
("Rapp_Error.mpr", "2010-12-02", "2010-12-02"),
("Ewe_Error.mpr", "2021-11-18", "2021-11-19"),
("col_27_issue_74.mpr", "2022-07-28", "2022-07-28"),
],
)
def test_MPR_dates(testdata_dir, filename, startdate, enddate):
"""Check that the start and end dates in .mpr files are read correctly."""
mpr = MPRfile(os.path.join(testdata_dir, filename))
assert mpr.startdate.strftime("%Y-%m-%d") == startdate
if enddate:
assert mpr.enddate.strftime("%Y-%m-%d") == enddate
else:
assert not hasattr(mpr, "enddate")
def test_open_MPR_fails_for_bad_file(testdata_dir):
with pytest.raises(ValueError, match="Invalid magic for .mpr file"):
MPRfile(os.path.join(testdata_dir, "arbin1.res"))
def timestamp_from_comments(comments):
for line in comments:
time_match = re.match(b"Acquisition started on : ([0-9/]+ [0-9:]+)", line)
if time_match:
timestamp = datetime.strptime(
time_match.group(1).decode("ascii"), "%m/%d/%Y %H:%M:%S"
)
return timestamp
raise AttributeError("No timestamp in comments")
def assert_MPR_matches_MPT(mpr, mpt, comments):
def assert_field_matches(fieldname, decimal):
if fieldname in mpr.dtype.fields:
assert_array_almost_equal(
mpr.data[fieldname], mpt[fieldname], decimal=decimal
)
def assert_field_exact(fieldname):
if fieldname in mpr.dtype.fields:
assert_array_equal(mpr.data[fieldname], mpt[fieldname])
assert_array_equal(mpr.get_flag("mode"), mpt["mode"])
assert_array_equal(mpr.get_flag("ox/red"), mpt["ox/red"])
assert_array_equal(mpr.get_flag("error"), mpt["error"])
assert_array_equal(mpr.get_flag("control changes"), mpt["control changes"])
if "Ns changes" in mpt.dtype.fields:
assert_array_equal(mpr.get_flag("Ns changes"), mpt["Ns changes"])
# Nothing uses the 0x40 bit of the flags
assert_array_equal(mpr.get_flag("counter inc."), mpt["counter inc."])
assert_array_almost_equal(
mpr.data["time/s"], mpt["time/s"], decimal=2
) # 2 digits in CSV
assert_field_matches("control/V/mA", decimal=6)
assert_field_matches("control/V", decimal=6)
assert_array_almost_equal(
mpr.data["Ewe/V"], mpt["Ewe/V"], decimal=6
) # 32 bit float precision
assert_field_matches("dQ/mA.h", decimal=16) # 64 bit float precision
assert_field_matches("P/W", decimal=10) # 32 bit float precision for 1.xxE-5
assert_field_matches("I/mA", decimal=6) # 32 bit float precision
assert_field_exact("cycle number")
assert_field_matches("(Q-Qo)/C", decimal=6) # 32 bit float precision
try:
assert timestamp_from_comments(comments) == mpr.timestamp.replace(microsecond=0)
except AttributeError:
pass
def assert_MPR_matches_MPT_v2(mpr, mpt, comments):
"""
Asserts that the fields in the MPR.data ar the same as in the MPT.
Modified from assert_MPR_matches_MPT. Automatically converts dtype from MPT data
to dtype from MPR data before comparing the columns.
Special case for EIS_indicators: these fields are valid only at f<100kHz so their
values are replaced by -1 or 0 at high frequency in the MPT file, this is not the
case in the MPR data.
Parameters
----------
mpr : MPRfile
Data extracted with the MPRfile class.
mpt : np.array
Data extracted with MPTfile method.
Returns
-------
None.
"""
def assert_field_matches(fieldname):
EIS_quality_indicators = [
"THD Ewe/%",
"NSD Ewe/%",
"NSR Ewe/%",
"|Ewe h2|/V",
"|Ewe h3|/V",
"|Ewe h4|/V",
"|Ewe h5|/V",
"|Ewe h6|/V",
"|Ewe h7|/V",
"THD I/%",
"NSD I/%",
"NSR I/%",
"|I h2|/A",
"|I h3|/A",
"|I h4|/A",
"|I h5|/A",
"|I h6|/A",
"|I h7|/A",
]
if fieldname in EIS_quality_indicators: # EIS quality indicators only valid for f < 100kHz
index_inf_100k = np.where(mpr.data["freq/Hz"] < 100000)[0]
assert_allclose(
mpr.data[index_inf_100k][fieldname],
mpt[index_inf_100k][fieldname].astype(mpr.data[fieldname].dtype),
)
elif fieldname == "<Ewe>/V":
assert_allclose(
mpr.data[fieldname],
mpt["Ewe/V"].astype(mpr.data[fieldname].dtype),
)
elif fieldname == "<I>/mA":
assert_allclose(
mpr.data[fieldname],
mpt["I/mA"].astype(mpr.data[fieldname].dtype),
)
elif fieldname == "dq/mA.h":
assert_allclose(
mpr.data[fieldname],
mpt["dQ/mA.h"].astype(mpr.data[fieldname].dtype),
)
else:
assert_allclose(
mpr.data[fieldname],
mpt[fieldname].astype(mpr.data[fieldname].dtype),
)
def assert_field_exact(fieldname):
if fieldname in mpr.dtype.fields:
assert_array_equal(mpr.data[fieldname], mpt[fieldname])
for key in mpr.flags_dict.keys():
assert_array_equal(mpr.get_flag(key), mpt[key])
for d in mpr.dtype.descr[1:]:
assert_field_matches(d[0])
try:
assert timestamp_from_comments(comments) == mpr.timestamp.replace(microsecond=0)
except AttributeError:
pass
@pytest.mark.parametrize(
"basename",
[
"bio_logic1",
"bio_logic2",
# No bio_logic3.mpt file
"bio_logic4",
# bio_logic5 and bio_logic6 are special cases
"CV_C01",
"121_CA_455nm_6V_30min_C01",
"020-formation_CB5",
],
)
def test_MPR_matches_MPT(testdata_dir, basename):
"""Check the MPR parser against the MPT parser.
Load a binary .mpr file and a text .mpt file which should contain
exactly the same data. Check that the loaded data actually match.
"""
binpath = os.path.join(testdata_dir, basename + ".mpr")
txtpath = os.path.join(testdata_dir, basename + ".mpt")
mpr = MPRfile(binpath)
mpt, comments = MPTfile(txtpath, encoding="latin1")
assert_MPR_matches_MPT(mpr, mpt, comments)
def test_MPR5_matches_MPT5(testdata_dir):
mpr = MPRfile(os.path.join(testdata_dir, "bio_logic5.mpr"))
mpt, comments = MPTfile(
(
re.sub(b"\tXXX\t", b"\t0\t", line)
for line in open(os.path.join(testdata_dir, "bio_logic5.mpt"), mode="rb")
)
)
assert_MPR_matches_MPT(mpr, mpt, comments)
def test_MPR6_matches_MPT6(testdata_dir):
mpr = MPRfile(os.path.join(testdata_dir, "bio_logic6.mpr"))
mpt, comments = MPTfile(os.path.join(testdata_dir, "bio_logic6.mpt"))
mpr.data = mpr.data[:958] # .mpt file is incomplete
assert_MPR_matches_MPT(mpr, mpt, comments)
@pytest.mark.parametrize(
"basename_v1150",
["v1150_CA", "v1150_CP", "v1150_GCPL", "v1150_GEIS", "v1150_MB", "v1150_OCV", "v1150_PEIS"],
)
def test_MPR_matches_MPT_v1150(testdata_dir, basename_v1150):
"""Check the MPR parser against the MPT parser.
Load a binary .mpr file and a text .mpt file which should contain
exactly the same data. Check that the loaded data actually match.
"""
binpath = os.path.join(testdata_dir, "v1150", basename_v1150 + ".mpr")
txtpath = os.path.join(testdata_dir, "v1150", basename_v1150 + ".mpt")
mpr = MPRfile(binpath)
mpt, comments = MPTfile(txtpath, encoding="latin1")
assert_MPR_matches_MPT_v2(mpr, mpt, comments)
@pytest.mark.skip(reason="Test data file is missing")
def test_loop_from_file(testdata_dir):
"""Check if the loop_index is correctly extracted from the _LOOP.txt file
"""
mpr = MPRfile(os.path.join(testdata_dir, "running", "running_OCV.mpr"))
assert mpr.loop_index is not None, "No loop_index found"
assert len(mpr.loop_index) == 4, "loop_index is not the right size"
assert_array_equal(mpr.loop_index, [0, 4, 8, 11], "loop_index values are wrong")
@pytest.mark.skip(reason="Test data file is missing")
def test_timestamp_from_file(testdata_dir):
"""Check if the loop_index is correctly extracted from the _LOOP.txt file
"""
mpr = MPRfile(os.path.join(testdata_dir, "running", "running_OCV.mpr"))
assert hasattr(mpr, "timestamp"), "No timestamp found"
assert mpr.timestamp.timestamp() == pytest.approx(1707299985.908), "timestamp value is wrong"