def test_svgp_model(self):
data = generate_data(sample_size=4, n_sites=2, randseed=3)
m = pynm.PyNM(data)
m.gp_normative_model(method='approx')
assert 'GP_pred' in m.data.columns
assert math.isclose(0, m.data['GP_residuals'].mean(), abs_tol=0.5)
def test_get_masks_all_CON(self):
a = np.array(list(range(12)))
data = generate_data(randseed=1)
m = pynm.PyNM(data)
ctr, prob = m._get_masks()
assert a[ctr].shape[0] == 12
assert a[prob].shape[0] == 0
def test_het_res(self):
data = dataset_het()
m = pynm.PyNM(data, score='score',conf='x',confounds = ['x'],group='train_sample')
with pytest.warns(Warning) as record:
m.gp_normative_model(method='exact')
assert len(record) == 1
assert record[0].message.args[0] == "The residuals are heteroskedastic!"
def test_get_masks(self):
a = np.array(list(range(6)))
data = generate_data(randseed=3)
m = pynm.PyNM(data)
ctr, prob = m._get_masks()
assert a[ctr].shape[0] == 5
assert a[prob][0] == 3
def test_use_approx_exact(self):
data = generate_data(randseed=3,sample_size=2000)
m = pynm.PyNM(data)
with pytest.warns(Warning) as record:
use_approx = m._use_approx(method='exact')
assert len(record) == 1
assert record[0].message.args[0] == "Exact GP model with over 2000 data points requires large amounts of time and memory, continuing with exact model."
assert use_approx == False
def test_get_conf_mat(self):
data = generate_data(randseed=3)
m = pynm.PyNM(data)
conf_mat = m._get_conf_mat()
assert conf_mat.shape[0] == 6
assert conf_mat.shape[1] == 3
for i in range(3):
assert not isinstance(conf_mat[0, i], str)
def test_svgp_init(self):
from pynm.approx import SVGP
data = generate_data(randseed=3)
m = pynm.PyNM(data)
conf_mat = m._get_conf_mat()
ctr,prob = m._get_masks()
score = m._get_score()
svgp = SVGP(conf_mat,score,ctr)
assert svgp.n_train == 5
assert svgp.n_test == 6
def test_svgp_train(self):
from pynm.approx import SVGP
data = generate_data(randseed=3)
m = pynm.PyNM(data)
conf_mat = m._get_conf_mat()
ctr,prob = m._get_masks()
score = m._get_score()
svgp = SVGP(conf_mat,score,ctr)
svgp.train(num_epochs = 2)
assert len(svgp.loss) == 2
def test_svgp_predict(self):
from pynm.approx import SVGP
data = generate_data(randseed=3)
m = pynm.PyNM(data)
conf_mat = m._get_conf_mat()
ctr,prob = m._get_masks()
score = m._get_score()
svgp = SVGP(conf_mat,score,ctr)
svgp.train(num_epochs = 2)
means,sigmas = svgp.predict()
assert means.size(0) == 6
assert sigmas.size(0) == 6
def main():
params = vars(_cli_parser())
confounds = params['confounds'].split(',')
data = pd.read_csv(params['pheno_p'])
m = pynm.PyNM(data, params['score'], params['group'], params['conf'],
confounds, params['train_sample'])
#Add a column to data w/ number controls used in this bin
m.bins_num()
#Run models
m.loess_normative_model()
m.centiles_normative_model()
m.gp_normative_model(length_scale=args.length_scale,
nu=args.nu,
method=args.method,
batch_size=args.batch_size,
n_inducing=args.n_inducing,
num_epochs=args.num_epochs)
m.data.to_csv(args.out_p, index=False)
def test_set_group_names_PROB_CON_all_CON(self):
data = generate_data(randseed=1)
m = pynm.PyNM(data)
assert m.CTR == 'CTR'
assert m.PROB == 'PROB'
def test_set_group_names_01(self):
data = generate_data(randseed=3, group='01')
m = pynm.PyNM(data)
assert m.CTR == 0
assert m.PROB == 1
def test_set_group_manual_zero_col(self):
data = generate_data(randseed=3, group='01')
data['train_sample'] = 0
with pytest.raises(ValueError):
m = pynm.PyNM(data,train_sample='manual')
def test_set_group_33(self):
data = generate_data(randseed=3, group='01')
m = pynm.PyNM(data,train_sample='0.33')
assert m.group == 'train_sample'
assert m.data['train_sample'].sum() == 1
assert m.data[(m.data['train_sample']==1) & (m.data['group']== 1)].shape[0] == 0
def test_create_bins(self):
data = generate_data(randseed=3)
m = pynm.PyNM(data,bin_spacing=8,bin_width=1.5)
m.centiles_normative_model()
assert m.bins is not None
def test_set_group_manual_good_col(self):
data = generate_data(randseed=3, group='01')
data['train_sample'] = [1,1,0,0,0,0]
m = pynm.PyNM(data,train_sample='manual')
assert m.PROB == 0
assert m.group == 'train_sample'
def test_homo_res(self):
data = dataset_homo()
m = pynm.PyNM(data, score='score',conf='x',confounds = ['x'],group='train_sample')
with pytest.warns(None) as record:
m.gp_normative_model(method='exact')
assert len(record) == 0
def test_bins_num(self):
data = generate_data(randseed=11)
m = pynm.PyNM(data,bin_spacing=5, bin_width=10)
m._create_bins()
assert len(m.bins) == 6
def test_loess_rank(self):
data = generate_data(randseed=11)
m = pynm.PyNM(data,bin_spacing=8,bin_width=1.5)
m.loess_normative_model()
assert np.sum(m.data.LOESS_rank) == 1
def test_plot(self):
data = generate_data(randseed=3)
m = pynm.PyNM(data)
m.gp_normative_model()
assert m.plot() is None
def test_use_approx_approx(self):
data = generate_data(randseed=3,sample_size=1000)
m = pynm.PyNM(data)
assert m._use_approx(method='approx') == True
def test_use_approx_auto_small(self):
data = generate_data(randseed=3)
m = pynm.PyNM(data)
assert m._use_approx(method='auto') == False
def test_set_group_controls(self):
data = generate_data(randseed=3, group='01')
m = pynm.PyNM(data,train_sample='controls')
assert m.group == 'group'
def test_loess_normative_model(self):
data = generate_data(randseed=11)
m = pynm.PyNM(data,bin_spacing=8,bin_width=1.5)
m.loess_normative_model()
assert math.isclose(2.3482, np.sum(m.data.LOESS_pred), abs_tol=0.00001)
def test_centiles_normative_model(self):
data = generate_data(randseed=11)
m = pynm.PyNM(data,bin_spacing=8,bin_width=1.5)
m.centiles_normative_model()
assert np.sum(m.data.Centiles) == 446
