Parsing the numerical output from Sensovation SensoSpot image analysis.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 

552 lines
15 KiB

""" Stub file for testing the project """
from pathlib import Path
import numpy
import pytest
EXAMPLE_DIR_WO_PARAMS = "mtp_wo_parameters"
EXAMPLE_DIR_WITH_PARAMS = "mtp_with_parameters"
@pytest.fixture
def example_dir(request):
root_dir = Path(request.config.rootdir)
yield root_dir / "example_data"
@pytest.fixture
def example_file(example_dir):
data_dir = example_dir / EXAMPLE_DIR_WO_PARAMS
yield data_dir / "160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv"
@pytest.fixture
def exposure_df():
from pandas import DataFrame
yield DataFrame(data={"Exposure.Id": [1, 2, 3]})
@pytest.fixture
def dir_for_caching(tmpdir, example_file):
import shutil
temp_path = Path(tmpdir)
dest = temp_path / example_file.name
shutil.copy(example_file, dest)
yield temp_path
@pytest.mark.parametrize(
"sub_dir, file_name",
[
(
EXAMPLE_DIR_WO_PARAMS,
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv",
),
(
EXAMPLE_DIR_WITH_PARAMS,
"160210_SG2-010-001_Regen_cy3100_1_A1_1.csv",
),
],
)
def test_parse_csv(example_dir, sub_dir, file_name):
from sensovation_data_parser.parser import _parse_csv
result = _parse_csv(example_dir / sub_dir / file_name)
columns = {
" ID ": numpy.int64,
"Pos.X": numpy.int64,
"Pos.Y": numpy.int64,
"Bkg.Mean": float,
"Spot.Mean": float,
"Bkg.Median": float,
"Spot.Median": float,
"Bkg.StdDev": float,
"Spot.StdDev": float,
"Bkg.Sum": numpy.int64,
"Spot.Sum": numpy.int64,
"Bkg.Area": numpy.int64,
"Spot.Area": numpy.int64,
"Spot.Sat. (%)": numpy.int64,
"Found": numpy.bool_,
"Pos.Nom.X": numpy.int64,
"Pos.Nom.Y": numpy.int64,
"Dia.": numpy.int64,
"Rect.": str,
"Contour": object, # ignore the type of contour
}
assert set(result.columns) == set(columns.keys())
assert len(result[" ID "].unique()) == 100
assert len(result) == 100
for column, value_type in columns.items():
assert isinstance(result[column][0], value_type)
def test_parse_csv_no_array(example_dir):
from sensovation_data_parser.parser import _parse_csv
result = _parse_csv(example_dir / "no_array_A1_1.csv")
assert len(result) == 1
assert result[" ID "][0] == 0
@pytest.mark.parametrize(
"input, expected", [("", "."), ("..,", "."), (".,,", ","), ("..,,", "."),]
)
def test_guess_decimal_separator_returns_correct_separator(input, expected):
from sensovation_data_parser.parser import _guess_decimal_separator
from io import StringIO
handle = StringIO(f"header\n{input}\n")
result = _guess_decimal_separator(handle)
assert result == expected
def test_guess_decimal_separator_rewinds_handle():
from sensovation_data_parser.parser import _guess_decimal_separator
from io import StringIO
handle = StringIO(f"header\n{input}\n")
_guess_decimal_separator(handle)
assert next(handle) == "header\n"
def test_well_regex_ok():
from sensovation_data_parser.parser import REGEX_WELL
result = REGEX_WELL.match("AbC123")
assert result["row"] == "AbC"
assert result["column"] == "123"
@pytest.mark.parametrize("input", ["", "A", "1", "1A", "-1", "A-"])
def test_well_regex_no_match(input):
from sensovation_data_parser.parser import REGEX_WELL
result = REGEX_WELL.match(input)
assert result is None
@pytest.mark.parametrize(
"filename, expected",
[("A1_1.csv", ("A", 1, 1)), ("test/measurement_1_H12_2", ("H", 12, 2)),],
)
def test_extract_measurement_info_ok(filename, expected):
from sensovation_data_parser.parser import _extract_measurement_info
result = _extract_measurement_info(filename)
assert result == expected
@pytest.mark.parametrize("filename", ["wrong_exposure_A1_B", "no_well_XX_1"])
def test_extract_measurement_info_raises_error(filename):
from sensovation_data_parser.parser import _extract_measurement_info
with pytest.raises(ValueError):
_extract_measurement_info(filename)
def test_cleanup_data_columns():
from sensovation_data_parser.parser import _cleanup_data_columns
from pandas import DataFrame
columns = ["Rect.", "Contour", " ID ", "Found", "Dia."]
data = {col: [i] for i, col in enumerate(columns)}
data_frame = DataFrame(data=data)
result = _cleanup_data_columns(data_frame)
assert set(result.columns) == {"Pos.Id", "Spot.Found", "Spot.Diameter"}
assert result["Pos.Id"][0] == 2
assert result["Spot.Found"][0] == 3
assert result["Spot.Diameter"][0] == 4
def test_parse_file(example_file):
from sensovation_data_parser.parser import parse_file
result = parse_file(example_file)
columns = {
"Pos.Id",
"Pos.X",
"Pos.Y",
"Bkg.Mean",
"Spot.Mean",
"Bkg.Median",
"Spot.Median",
"Bkg.StdDev",
"Spot.StdDev",
"Bkg.Sum",
"Spot.Sum",
"Bkg.Area",
"Spot.Area",
"Spot.Sat. (%)",
"Spot.Found",
"Pos.Nom.X",
"Pos.Nom.Y",
"Spot.Diameter",
"Field.Row",
"Field.Column",
"Exposure.Id",
}
assert set(result.columns) == columns
assert result["Field.Row"][0] == "A"
assert result["Field.Column"][0] == 1
assert result["Exposure.Id"][0] == 1
@pytest.mark.parametrize(
"file_list",
[
[
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv",
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_2.csv",
],
["160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv"],
],
)
def testparse_multiple_files_ok(example_dir, file_list):
from sensovation_data_parser.parser import parse_multiple_files
sub_dir = example_dir / EXAMPLE_DIR_WO_PARAMS
files = [sub_dir / file for file in file_list]
data_frame = parse_multiple_files(files)
print(data_frame["Exposure.Id"].unique())
assert len(data_frame) == 100 * len(files)
assert len(data_frame["Exposure.Id"].unique()) == len(files)
def testparse_multiple_files_empty_file_list():
from sensovation_data_parser.parser import parse_multiple_files
with pytest.raises(ValueError):
parse_multiple_files([])
def testparse_multiple_files_empty_array(example_dir):
from sensovation_data_parser.parser import parse_multiple_files
files = [example_dir / "no_array_A1_1.csv"]
data_frame = parse_multiple_files(files)
print(data_frame["Exposure.Id"].unique())
assert len(data_frame) == 1
def test_list_csv_files(example_dir):
from sensovation_data_parser.parser import _list_csv_files
result = list(_list_csv_files(example_dir / EXAMPLE_DIR_WITH_PARAMS))
assert len(result) == 36 * 3
assert all(str(item).endswith(".csv") for item in result)
assert all(not item.stem.startswith(".") for item in result)
def test_parse_folder(example_dir):
from sensovation_data_parser.parser import parse_folder
data_frame = parse_folder(example_dir / EXAMPLE_DIR_WITH_PARAMS)
assert len(data_frame) == 36 * 3 * 100
assert len(data_frame["Field.Row"].unique()) == 3
assert len(data_frame["Field.Column"].unique()) == 12
assert len(data_frame["Exposure.Id"].unique()) == 3
assert len(data_frame["Pos.Id"].unique()) == 100
def test_sanity_check_ok(example_dir):
from sensovation_data_parser.parser import (
_sanity_check,
parse_multiple_files,
)
sub_dir = example_dir / EXAMPLE_DIR_WO_PARAMS
file_list = [
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv",
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_2.csv",
]
files = [sub_dir / file for file in file_list]
data_frame = parse_multiple_files(files)
result = _sanity_check(data_frame)
assert len(result) == len(data_frame)
def test_sanity_check_raises_value_error(example_dir):
from sensovation_data_parser.parser import (
_sanity_check,
parse_multiple_files,
)
sub_dir = example_dir / EXAMPLE_DIR_WO_PARAMS
file_list = [
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_1.csv",
"160218_SG2-013-001_Regen1_Cy3-100_1_A1_2.csv",
]
files = [sub_dir / file for file in file_list]
data_frame = parse_multiple_files(files)
data_frame = data_frame.drop(data_frame.index[1])
with pytest.raises(ValueError):
_sanity_check(data_frame)
def test_search_channel_info_file_ok(example_dir):
from sensovation_data_parser.parser import _search_channel_info_file
result = _search_channel_info_file(example_dir / EXAMPLE_DIR_WITH_PARAMS)
assert result.suffix == ".svexp"
def test_search_channel_info_file_no_parameters_folder(example_dir):
from sensovation_data_parser.parser import _search_channel_info_file
result = _search_channel_info_file(example_dir / EXAMPLE_DIR_WO_PARAMS)
assert result is None
def test_search_channel_info_file_no_parameters_file(tmpdir):
from sensovation_data_parser.parser import _search_channel_info_file
params_dir = tmpdir / "Parameters"
params_dir.mkdir()
result = _search_channel_info_file(tmpdir)
assert result is None
def test_parse_channel_info(example_dir):
from sensovation_data_parser.parser import (
_search_channel_info_file,
_parse_channel_info,
)
params = _search_channel_info_file(example_dir / EXAMPLE_DIR_WITH_PARAMS)
result = _parse_channel_info(params)
assert set(result.keys()) == {1, 2, 3}
assert result[1] == ("green", 100)
assert result[2] == ("red", 150)
assert result[3] == ("red", 15)
def test_get_valid_exposure_map_provided_ok(exposure_df):
from sensovation_data_parser.parser import (
_get_valid_exposure_map,
ExposureInfo,
)
dummy_value = ExposureInfo(None, None)
exposure_map = {1: dummy_value, 2: dummy_value, 3: dummy_value}
result = _get_valid_exposure_map(
"/nonexistent", exposure_df, exposure_map=exposure_map
)
assert result == exposure_map
def test_get_valid_exposure_map_provided_not_ok(exposure_df):
from sensovation_data_parser.parser import _get_valid_exposure_map
exposure_map = {1: None, 2: None}
result = _get_valid_exposure_map(
"/nonexistent", exposure_df, exposure_map=exposure_map
)
assert set(result.keys()) == {1, 2, 3}
assert all(v == (None, None) for v in result.values())
def test_get_valid_exposure_map_info_from_file_ok(example_dir, exposure_df):
from sensovation_data_parser.parser import _get_valid_exposure_map
result = _get_valid_exposure_map(
example_dir / EXAMPLE_DIR_WITH_PARAMS, exposure_df, exposure_map=None
)
assert set(result.keys()) == {1, 2, 3}
assert result[1] == ("green", 100)
assert result[2] == ("red", 150)
assert result[3] == ("red", 15)
def test_get_valid_exposure_map_info_from_file_not_ok(
example_dir, exposure_df
):
from sensovation_data_parser.parser import _get_valid_exposure_map
data_frame = exposure_df.drop(exposure_df.index[1])
result = _get_valid_exposure_map(
example_dir / EXAMPLE_DIR_WITH_PARAMS, data_frame, exposure_map=None
)
assert set(result.keys()) == {1, 3}
assert all(v == (None, None) for v in result.values())
def test_augment_exposure_map(exposure_df):
from sensovation_data_parser.parser import (
_augment_exposure_map,
ExposureInfo,
)
exposure_map = {
1: ExposureInfo("red", 10),
2: ExposureInfo("green", 20),
3: ExposureInfo("blue", 50),
}
result = _augment_exposure_map(exposure_df, exposure_map)
assert result["Exposure.Id"][0] == 1
assert result["Exposure.Channel"][0] == "red"
assert result["Exposure.Time"][0] == 10
assert result["Exposure.Id"][1] == 2
assert result["Exposure.Channel"][1] == "green"
assert result["Exposure.Time"][1] == 20
assert result["Exposure.Id"][2] == 3
assert result["Exposure.Channel"][2] == "blue"
assert result["Exposure.Time"][2] == 50
def test_process_folder_with_exposure_map(example_dir):
from sensovation_data_parser.parser import _process_folder
result = _process_folder(example_dir / EXAMPLE_DIR_WITH_PARAMS)
assert len(result) == 36 * 100 * 3
expected = [(1, "green", 100), (2, "red", 150), (3, "red", 15)]
for exposure_id, channel, time in expected:
mask = result["Exposure.Id"] == exposure_id
example_row = result.loc[mask].iloc[1]
assert example_row["Exposure.Channel"] == channel
assert example_row["Exposure.Time"] == time
def test_process_folder_without_exposure_map(example_dir):
from sensovation_data_parser.parser import _process_folder
from pandas import isnull
result = _process_folder(example_dir / EXAMPLE_DIR_WO_PARAMS)
assert len(result) == 96 * 100 * 3
for exposure_id in range(1, 4):
mask = result["Exposure.Id"] == exposure_id
example_row = result.loc[mask].iloc[1]
print(type(example_row["Exposure.Channel"]))
assert isnull(example_row["Exposure.Channel"])
assert isnull(example_row["Exposure.Time"])
def test_process_folder_creates_cache(dir_for_caching):
from sensovation_data_parser.parser import (
process_folder,
CACHE_FILE_NAME,
)
cache_path = dir_for_caching / CACHE_FILE_NAME
assert not cache_path.is_file()
result = process_folder(dir_for_caching)
assert len(result) == 100
assert cache_path.is_file()
def test_process_folder_reads_from_cache(dir_for_caching, example_file):
from sensovation_data_parser.parser import process_folder
process_folder(dir_for_caching)
csv_file = dir_for_caching / example_file.name
csv_file.unlink()
result = process_folder(dir_for_caching)
assert len(result) == 100
def test_process_folder_read_cache_fails_silently(
dir_for_caching, exposure_df
):
from sensovation_data_parser.parser import (
process_folder,
CACHE_FILE_NAME,
)
cache_path = dir_for_caching / CACHE_FILE_NAME
exposure_df.to_hdf(cache_path, "unknown table")
result = process_folder(dir_for_caching)
assert result["Field.Row"][0] == "A"
def test_get_cache_table_name():
from sensovation_data_parser.parser import _get_cache_table_name
from sensovation_data_parser import __version__
result = _get_cache_table_name()
assert result.startswith("v")
assert result[1:] == __version__
def test_process_folder_read_cache_no_cache_arg(dir_for_caching, exposure_df):
from sensovation_data_parser.parser import (
process_folder,
_get_cache_table_name,
CACHE_FILE_NAME,
)
cache_path = dir_for_caching / CACHE_FILE_NAME
exposure_df.to_hdf(cache_path, _get_cache_table_name())
result = process_folder(dir_for_caching, use_cache=False)
assert result["Field.Row"][0] == "A"
def test_process_folder_writes_cache(dir_for_caching):
from sensovation_data_parser.parser import (
process_folder,
CACHE_FILE_NAME,
)
process_folder(dir_for_caching, use_cache=True)
cache_path = dir_for_caching / CACHE_FILE_NAME
assert cache_path.is_file()
def test_process_folder_writes_cache_no_cache_arg(dir_for_caching):
from sensovation_data_parser.parser import process_folder, CACHE_FILE_NAME
process_folder(dir_for_caching, use_cache=False)
cache_path = dir_for_caching / CACHE_FILE_NAME
assert not cache_path.is_file()