def test_single_censored_transfers_to_masked_on_setting():
va = Record()
assert not va.masked
va.censored = 1
assert 1 in va.masked
assert va.censored == {1}
def test_set_censored_multiple_times():
va = Record()
va.censored = 2
assert va.censored == {2}
assert 2 in va.masked
va.censored = 3
assert va.censored == {3}
assert 3 in va.masked
def test_mask_ranks(prep, censored, scores, ranks, expected):
prep.cause_list = scores.keys()
va = Record(scores=scores, censored=censored)
va.ranks = ranks
uniform = range(1000) # just needs length
cutoffs = dict(zip(scores.keys(), [99999] * len(scores)))
demog_restrictions = {}
lowest_rank = 9999 # value set if restricted
uniform_list_pos = 999
min_cause_score = defaultdict(lambda: 0)
prep.mask_ranks([va], uniform, cutoffs, demog_restrictions, lowest_rank,
uniform_list_pos, min_cause_score)
assert va.ranks == expected
def test_csmf_summed_to_one(prep):
causes = ['a', 'b', 'c']
counts = np.random.randint(10, 100, 3)
user_data = [Record('sid{}'.format(i), age=35, sex=i % 2 + 1, cause=cause)
for i, cause in enumerate(causes) for i in range(counts[i])]
csmf, csmf_by_sex = prep.calculate_csmf(user_data, [])
assert np.allclose(sum(csmf.values()), 1)
for sex, csmf_data in csmf_by_sex.items():
assert np.allclose(sum(csmf_data.values()), 1)
def test_generate_cause_rankings(prep):
prep.cause_list = [1]
train_scores = [
100, # rank 1
50, # rank 2
15, # rank 3 (triplicate)
15, # rank 4 (triplicate)
15, # rank 5 (triplicate)
10, # rank 6 (duplicate positive)
10, # rank 7 (duplicate positive)
-2, # rank 8 (duplicate negative)
-2, # rank 9 (duplicate negative)
-3, # rank 10
]
train_data = [Record(scores={1: s}) for s in train_scores]
uniform_scores = {1: np.sort([va.scores[1] for va in train_data])}
# Score, Rank within training
tests = [
(110, 0.5), # above highest score in train data
(100, 1), # at highest score in train data
(90, 1.5),
(50, 2), # at a value which exists in the train data
(15, 4), # at triplicate value in the train data
(11, 5.5),
(10, 6.5), # at a duplicate value in the train data
(0, 7.5), # zero (just in case)
(-1, 7.5), # negative value in range of train scores max to min
(-3, 10), # at lowest score in train data
(-5, 10.5), # below lowest score in train data
]
test_data = [Record(scores={1: score}) for score, rank in tests]
# Modifies list of Records in place and doesn't return anything
prep.generate_cause_rankings(test_data, uniform_scores)
predicted_test_ranks = [va.ranks[1] for va in test_data]
actual_test_ranks = [float(rank) for score, rank in tests]
assert predicted_test_ranks == actual_test_ranks
def test_csmf_sex_undetermined_unknown_age(prep):
"""
Redistributed CSMFs for undetermined causes of death should not include
biologically impossible causes for males and females.
"""
male_drops = []
female_drops = []
# remove sex specific drops
male_drops.extend(prep.data_module.FEMALE_CAUSES)
male_drops.extend(prep.data_module.MATERNAL_CAUSES)
female_drops.extend(prep.data_module.MALE_CAUSES)
# male and female user data with undetermined cause and age zero (unknown age)
user_data_male = [
Record('sid{}'.format(i), age=0, sex=1, cause34_name='Undetermined')
for i in range(7)
]
user_data_female = [
Record('sid{}'.format(i), age=0, sex=2, cause34_name='Undetermined')
for i in range(7)
]
user_data_unknown = [
Record('sid{}'.format(i), age=0, sex=0, cause34_name='Undetermined')
for i in range(7)
]
user_data = user_data_male + user_data_female + user_data_unknown
undetermined_weights = prep._get_undetermined_matrix()
csmf, csmf_by_sex = prep.calculate_csmf(user_data, undetermined_weights)
for sex, csmf_data in csmf_by_sex.items():
# check female causes not in male csmfs and vice versa
if sex == 1:
assert len(list(set(csmf_data.keys()) & set(male_drops))) == 0
if sex == 2:
assert len(list(set(csmf_data.keys()) & set(female_drops))) == 0
def test_csmf_summed_to_one(prep, malaria, hiv):
prep.malaria_region = malaria
prep.hiv_region = hiv
causes = prep.data_module.CAUSES.values()
user_data = [Record('sid{}'.format(i), age=.1, sex=i % 2 + 1, cause=cause)
for i in range(7) for cause in causes]
undetermined_weights = prep._get_undetermined_matrix()
csmf, csmf_by_sex = prep.calculate_csmf(user_data, undetermined_weights)
assert np.allclose(sum(csmf.values()), 1)
for sex, csmf_data in csmf_by_sex.items():
assert np.allclose(sum(csmf_data.values()), 1)
def test_csmf_sex_undetermined_plot(prep):
"""
Redistributed CSMFs for undetermined causes of death should not include
biologically impossible causes for males and females. Check that these
causes are not included in the CSMF figures.
"""
# male and female user data with undetermined cause and age zero (unknown age)
user_data_male = [Record('sid{}'.format(i), age=0, sex=1, cause34_name='Undetermined')
for i in range(7)]
user_data_female = [Record('sid{}'.format(i), age=0, sex=2, cause34_name='Undetermined')
for i in range(7)]
user_data_unknown = [Record('sid{}'.format(i), age=0, sex=0, cause34_name='Undetermined')
for i in range(7)]
user_data = user_data_male + user_data_female + user_data_unknown
undetermined_weights = prep._get_undetermined_matrix()
csmf, csmf_by_sex = prep.calculate_csmf(user_data, undetermined_weights)
for sex in range(1, 2):
# input graph data
graph_data = csmf_by_sex[sex]
cause_keys = graph_data.keys()
cause_fractions = graph_data.values()
#graph_title = module_key.capitalize() + ' CSMF' # not neessary to have the graph title
# graph_filename = graph_title.replace(' ', '-').lower() # not neccesary to have filename
max_value = max(cause_fractions)
xlocations = np.arange(len(cause_keys)) # the x locations for the groups
bar_width = .75 # the width of the bars
# Interactive mode off.
plt.ioff()
fig, ax = plt.subplots()
ax.set_ylabel('Mortality fractions')
ax.yaxis.grid()
ax.set_xticklabels(cause_keys)
ax.set_xticks(xlocations)
bar_width = .75 # the width of the bars
# Interactive mode off.
plt.ioff()
fig, ax = plt.subplots()
#ax.set_title(graph_title) # not neessary to have the graph title
ax.set_ylabel('Mortality fractions')
ax.yaxis.grid()
ax.set_xticklabels(cause_keys, rotation=90)
ax.set_xticks(xlocations)
ax.bar(xlocations, cause_fractions, bar_width, color='#C44440', align='center')
# Add whitespace at top of bar.
ax.set_ylim(top=max_value + max_value * 0.1)
# Add whitespace before first bar and after last.
plt.xlim([min(xlocations) - .5, max(xlocations) + 1.0])
# Add some spacing for rotated xlabels.
plt.subplots_adjust(bottom=0.60)
# neccessary to get acutally plot to get x-axis labels
fig.canvas.draw()
# list of x axis labels
labels = [item.get_text() for item in ax.get_xticklabels()]
# check biologically impossible causes
if sex == 1:
assert "Stroke" in labels
assert "Prostate Cancer" in labels
assert "Maternal" not in labels
if sex == 2:
assert "Stroke" in labels
assert "Maternal" in labels
assert "Cervical Cancer" in labels
assert "Prostate Cancer" not in labels
def test_reporter():
va = Record()
print va
def test_multiple_censored_transfers_to_masked_on_setting():
va = Record()
va.censored = [0, 1, 2, 3]
for x in range(4):
assert x in va.masked
assert va.censored == {0, 1, 2, 3}
def test_censored_transfers_to_masked_on_init():
va = Record(censored=(2, 3))
assert 2 in va.masked
assert 3 in va.masked
uniform = range(1000) # just needs length
cutoffs = dict(zip(scores.keys(), [99999] * len(scores)))
demog_restrictions = {}
lowest_rank = 9999 # value set if restricted
uniform_list_pos = 999
min_cause_score = defaultdict(lambda: 0)
prep.mask_ranks([va], uniform, cutoffs, demog_restrictions, lowest_rank,
uniform_list_pos, min_cause_score)
assert va.ranks == expected
@pytest.mark.parametrize('va, cause, cause_name', [
(Record(sid='rules', cause=1), 11, 'c11'),
(Record(sid='rules2', cause=2), 12, 'c12'),
(Record(sid='ranks', ranks={1: 1, 2: 2}), 11, 'c11'),
(Record(sid='rules_bad', cause='x', ranks={1: 1, 2: 2}), 11, 'c11'),
(Record(sid='tie', ranks={1: 1, 2: 1, 3: 2}), 11, 'c11'),
(Record(sid='lowest', ranks={1: 999, 2: 999, 3: 999}), None,
'Undetermined'),
])
def test_predict_with_rule(prep, va, cause, cause_name):
user_data = [va]
cause_reduction = {1: 11, 2: 12, 3: 13, 4: 14}
names34 = {11: 'c11', 12: 'c12', 13: 'c13', 14: 'c14'}
names46 = {1: 'c1', 2: 'c2', 3: 'c3', 4: 'c4'}
prep.predict(user_data, 999, cause_reduction, names34, names46)
assert va.cause34 == cause