def test_filter_features_libsvm_output_argparse():
"""
Make sure filter_features exits when passing in output libsvm files
"""
ff_cmd_args = ['foo.csv', 'bar.libsvm', '-f', 'a', 'b', 'c']
ff.main(argv=ff_cmd_args)
def test_filter_features_unknown_input_format():
"""
Make sure that filter_features exits when passing in an unknown input file format
"""
ff_cmd_args = ['foo.xxx', 'bar.csv', '-f', 'a', 'b', 'c']
ff.main(argv=ff_cmd_args)
def check_filter_features_raises_system_exit(cmd_args):
"""
Little helper to make test output cleaner for tests that check that
filter_features exits with the specified arguments.
"""
ff.main(cmd_args)
def check_filter_features_arff_argparse(filter_type, label_col='y',
id_col='id', inverse=False,
quiet=False):
"""
A utility function to check that we are setting up argument parsing
correctly for filter_features for ARFF file types. We are not checking
whether the results are correct because we have separate tests for that.
"""
# replace the run_configuration function that's called
# by the main() in filter_feature with a mocked up version
writer_class = skll.data.writers.ARFFWriter
# create some dummy input and output filenames
infile = join(_my_dir, 'other', 'test_filter_features_input.arff')
outfile = 'bar.arff'
# create a simple featureset with actual ids, labels and features
fs, _ = make_classification_data(num_labels=3, train_test_ratio=1.0)
writer = writer_class(infile, fs, label_col=label_col, id_col=id_col)
writer.write()
ff_cmd_args = [infile, outfile]
if filter_type == 'feature':
if inverse:
features_to_keep = ['f01', 'f04', 'f07', 'f10']
else:
features_to_keep = ['f02', 'f03', 'f05', 'f06', 'f08', 'f09']
ff_cmd_args.append('-f')
for f in features_to_keep:
ff_cmd_args.append(f)
elif filter_type == 'id':
if inverse:
ids_to_keep = ['EXAMPLE_{}'.format(x) for x in range(1, 100, 2)]
else:
ids_to_keep = ['EXAMPLE_{}'.format(x) for x in range(2, 102, 2)]
ff_cmd_args.append('-I')
for idee in ids_to_keep:
ff_cmd_args.append(idee)
elif filter_type == 'label':
if inverse:
labels_to_keep = ['0', '1']
else:
labels_to_keep = ['2']
ff_cmd_args.append('-L')
for lbl in labels_to_keep:
ff_cmd_args.append(lbl)
ff_cmd_args.extend(['-l', label_col])
ff_cmd_args.extend(['--id_col', id_col])
if inverse:
ff_cmd_args.append('-i')
if quiet:
ff_cmd_args.append('-q')
# Substitute mock methods for the main methods that get called by
# filter_features for arff files: FeatureSet.filter() and the __init__()
# method of the appropriate writer. We also need to mock the write()
# method to prevent actual writing.
with patch.object(FeatureSet, 'filter', autospec=True) as filter_mock, \
patch.object(writer_class, '__init__', autospec=True,
return_value=None) as write_init_mock, \
patch.object(writer_class, 'write', autospec=True) as write_mock:
ff.main(argv=ff_cmd_args)
# get the various arguments from the three mocked up methods
filter_pos_arguments, filter_kw_arguments = filter_mock.call_args
write_pos_arguments, write_kw_arguments = write_init_mock.call_args
# make sure that the arguments they got were the ones we specified
eq_(write_pos_arguments[1], outfile)
eq_(write_kw_arguments['quiet'], quiet)
# note that we cannot test the label_col column for the writer
# the reason is that is set conditionally and those conditions
# do not execute with mocking
eq_(filter_pos_arguments[0], fs)
eq_(filter_kw_arguments['inverse'], inverse)
if filter_type == 'feature':
eq_(filter_kw_arguments['features'], features_to_keep)
elif filter_type == 'id':
eq_(filter_kw_arguments['ids'], ids_to_keep)
elif filter_type == 'label':
eq_(filter_kw_arguments['labels'], labels_to_keep)
