def test_ttest_rel():
"""Test test.TTestRel()"""
ds = datasets.get_uv()
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds)
i1 = ds.eval("logical_and(A=='a1', B=='b1')")
i2 = ds.eval("logical_and(A=='a2', B=='b1')")
a1 = ds[i1, 'fltvar'].x
a2 = ds[i2, 'fltvar'].x
diff = a1 - a2
t, p = scipy.stats.ttest_rel(a1, a2)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, 1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t > 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, -1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, -1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t < 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
def test_ttest_rel():
"""Test test.TTestRel()"""
ds = datasets.get_uv()
res = test.TTestRel("fltvar", "A", "a1", "a2", "rm", "B=='b1'", ds)
i1 = ds.eval("logical_and(A=='a1', B=='b1')")
i2 = ds.eval("logical_and(A=='a2', B=='b1')")
a1 = ds[i1, "fltvar"].x
a2 = ds[i2, "fltvar"].x
diff = a1 - a2
t, p = scipy.stats.ttest_rel(a1, a2)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel("fltvar", "A", "a1", "a2", "rm", "B=='b1'", ds, 1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t > 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel("fltvar", "A", "a1", "a2", "rm", "B=='b1'", ds, -1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, -1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t < 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
def source_trfs(
ress: ResultCollection,
heading: FMTextArg = None,
brain_view: Union[str, Sequence[float]] = None,
axw: float = None,
surf: str = 'inflated',
cortex: Any = ((1.00,) * 3, (.4,) * 3),
vmax: float = None,
xlim: Tuple[float, float] = None,
times: Sequence[float] = None,
cmap: str = None,
labels: Dict[str, str] = None,
rasterize: bool = None,
brain_timewindow: float = 0.050
):
"Only used for TRFExperiment model-test"
layout = BrainLayout(brain_view, axw)
dt = brain_timewindow / 2
if heading is not None:
doc = fmtxt.Section(heading)
else:
doc = fmtxt.FMText()
if cmap is None:
cmap = 'lux-a'
if labels is None:
labels = {}
trf_table = fmtxt.Table('ll')
for key, res in ress.items():
trf_resampled = resample(res.masked_difference(), 1000)
label = labels.get(key, key)
if rasterize is None:
rasterize = len(trf_resampled.source) > 500
# times for anatomical plots
if times is None:
trf_tc = abs(trf_resampled).sum('source')
trf_tc_mask = (~trf_resampled.get_mask()).sum('source') >= 10
times_ = find_peak_times(trf_tc, trf_tc_mask)
else:
times_ = times
# butterfly-plot
p = plot.Butterfly(trf_resampled, h=3, w=4, ylabel=False, title=label, vmax=vmax, xlim=xlim, show=False)
for t in times_:
p.add_vline(t, color='k')
trf_table.cell(fmtxt.asfmtext(p, rasterize=rasterize))
p.close()
# peak sources
if not times_:
trf_table.cell()
sp = plot.brain.SequencePlotter()
if layout.brain_view:
sp.set_parallel_view(*layout.brain_view)
sp.set_brain_args(surf=surf, cortex=cortex)
for t in times_:
yt = trf_resampled.mean(time=(t - dt, t + dt + 0.001))
if isinstance(cmap, str):
vmax_ = vmax or max(-yt.min(), yt.max()) or 1
cmap_ = plot.soft_threshold_colormap(cmap, vmax_ / 10, vmax_)
else:
cmap_ = cmap
sp.add_ndvar(yt, cmap=cmap_, label=f'{t * 1000:.0f} ms', smoothing_steps=10)
p = sp.plot_table(view='lateral', orientation='vertical', **layout.table_args)
trf_table.cell(p)
p.close()
doc.append(fmtxt.Figure(trf_table))
return doc
def test_ttest():
"""Test univariate t-test functions"""
ds = datasets.get_uv()
print(test.ttest('fltvar', ds=ds))
print(test.ttest('fltvar', 'A', ds=ds))
print(test.ttest('fltvar', 'A%B', ds=ds))
print(test.ttest('fltvar', 'A', match='rm', ds=ds))
print(test.ttest('fltvar', 'A', 'a1', match='rm', ds=ds))
print(test.ttest('fltvar', 'A%B', ('a1', 'b1'), match='rm', ds=ds))
# Prepare data for scipy
a1_index = ds.eval("A == 'a1'")
a2_index = ds.eval("A == 'a2'")
b1_index = ds.eval("B == 'b1'")
a1_in_b1_index = np.logical_and(a1_index, b1_index)
a2_in_b1_index = np.logical_and(a2_index, b1_index)
# TTest1Samp
res = test.TTest1Sample('fltvar', ds=ds)
t, p = scipy.stats.ttest_1samp(ds['fltvar'], 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
res = test.TTest1Sample('fltvar', ds=ds, tail=1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2., 10)
# TTestInd
res = test.TTestInd('fltvar', 'A', 'a1', 'a2', ds=ds)
t, p = scipy.stats.ttest_ind(ds[a1_index, 'fltvar'], ds[a2_index, 'fltvar'])
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
# TTestRel
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds)
a1 = ds[a1_in_b1_index, 'fltvar'].x
a2 = ds[a2_in_b1_index, 'fltvar'].x
difference = a1 - a2
t, p = scipy.stats.ttest_rel(a1, a2)
assert_array_equal(res.difference.x, difference)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, 1)
assert_array_equal(res.difference.x, difference)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t > 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, -1)
assert_array_equal(res.difference.x, difference)
eq_(res.df, len(a1) - 1)
eq_(res.tail, -1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t < 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
# alternative argspec
a1 = ds.eval("fltvar[(B == 'b1') & (A == 'a1')]")
a2 = ds.eval("fltvar[(B == 'b1') & (A == 'a2')]")
res_alt = test.TTestRel(a1, a2, tail=-1)
print(res_alt)
assert res_alt.t == res.t
assert res_alt.p == res.p
def test_anova():
"Test testnd.anova()"
ds = datasets.get_uts(True, nrm=True)
testnd.anova('utsnd', 'A*B', ds=ds)
for samples in (0, 2):
logging.info("TEST: samples=%r" % samples)
testnd.anova('utsnd', 'A*B', ds=ds, samples=samples)
testnd.anova('utsnd', 'A*B', ds=ds, samples=samples, pmin=0.05)
res = testnd.anova('utsnd', 'A*B', ds=ds, samples=samples, tfce=True)
assert res._plot_model() == 'A%B'
asfmtext(res)
res = testnd.anova('utsnd', 'A*B*rm', match=False, ds=ds, samples=0, pmin=0.05)
assert repr(res) == "<anova 'utsnd', 'A*B*rm', match=False, samples=0, pmin=0.05, 'A': 17 clusters, 'B': 20 clusters, 'A x B': 22 clusters>"
assert res._plot_model() == 'A%B'
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=2, pmin=0.05)
assert res.match == 'rm'
assert repr(res) == "<anova 'utsnd', 'A*B*rm', match='rm', samples=2, pmin=0.05, 'A': 17 clusters, p < .001, 'B': 20 clusters, p < .001, 'A x B': 22 clusters, p < .001>"
assert res._plot_model() == 'A%B'
# persistence
string = pickle.dumps(res, protocol=pickle.HIGHEST_PROTOCOL)
res_ = pickle.loads(string)
assert repr(res_) == repr(res)
assert res_._plot_model() == 'A%B'
# threshold-free
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=10)
assert res.match == 'rm'
assert repr(res) == "<anova 'utsnd', 'A*B*rm', match='rm', samples=10, 'A': p < .001, 'B': p < .001, 'A x B': p < .001>"
assert 'A clusters' in res.clusters.info
assert 'B clusters' in res.clusters.info
assert 'A x B clusters' in res.clusters.info
# no clusters
res = testnd.anova('uts', 'B', sub="A=='a1'", ds=ds, samples=5, pmin=0.05, mintime=0.02)
repr(res)
assert 'v' in res.clusters
assert 'p' in res.clusters
assert res._plot_model() == 'B'
# all effects with clusters
res = testnd.anova('uts', 'A*B*rm', match=False, ds=ds, samples=5, pmin=0.05, tstart=0.1, mintime=0.02)
assert set(res.clusters['effect'].cells) == set(res.effects)
# some effects with clusters, some without
res = testnd.anova('uts', 'A*B*rm', ds=ds, samples=5, pmin=0.05, tstart=0.37, mintime=0.02)
assert res.match == 'rm'
string = pickle.dumps(res, pickle.HIGHEST_PROTOCOL)
res_ = pickle.loads(string)
assert_dataobj_equal(res.clusters, res_.clusters)
# test multi-effect results (with persistence)
# UTS
res = testnd.anova('uts', 'A*B*rm', ds=ds, samples=5)
assert res.match == 'rm'
repr(res)
string = pickle.dumps(res, pickle.HIGHEST_PROTOCOL)
resr = pickle.loads(string)
tf_clusters = resr.find_clusters(pmin=0.05)
peaks = resr.find_peaks()
assert_dataobj_equal(tf_clusters, res.find_clusters(pmin=0.05))
assert_dataobj_equal(peaks, res.find_peaks())
assert tf_clusters.eval("p.min()") == peaks.eval("p.min()")
unmasked = resr.f[0]
masked = resr.masked_parameter_map(effect=0, pmin=0.05)
assert_array_equal(masked.x <= unmasked.x, True)
# reproducibility
decimal = 12 if IS_WINDOWS else None # FIXME: why is Windows sometimes different???
res0 = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
assert_dataset_equal(res.clusters, res0.clusters, decimal=decimal)
configure(n_workers=0)
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
assert_dataset_equal(res.clusters, res0.clusters, decimal=decimal)
configure(n_workers=True)
# permutation
eelbrain._stats.permutation._YIELD_ORIGINAL = 1
samples = 4
# raw
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=samples)
for dist in res._cdist:
assert len(dist.dist) == samples
assert_array_equal(dist.dist, dist.parameter_map.abs().max())
# TFCE
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, tfce=True, samples=samples)
for dist in res._cdist:
assert len(dist.dist) == samples
assert_array_equal(dist.dist, dist.tfce_map.abs().max())
# thresholded
res1 = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=samples)
clusters = res1.find_clusters()
for dist, effect in zip(res1._cdist, res1.effects):
effect_idx = clusters.eval("effect == %r" % effect)
vmax = clusters[effect_idx, 'v'].abs().max()
assert len(dist.dist) == samples
assert_array_equal(dist.dist, vmax)
eelbrain._stats.permutation._YIELD_ORIGINAL = 0
# 1d TFCE
configure(n_workers=0)
res = testnd.anova('utsnd.rms(time=(0.1, 0.3))', 'A*B*rm', ds=ds, tfce=True, samples=samples)
configure(n_workers=True)
# zero variance
res2 = testnd.anova('utsnd', 'A', ds=ds)
ds['utsnd'].x[:, 1, 10] = 0.
zero_var = ds['utsnd'].var('case') == 0
zv_index = tuple(i[0] for i in zero_var.nonzero())
res1_zv = testnd.anova('utsnd', 'A*B*rm', ds=ds)
res2_zv = testnd.anova('utsnd', 'A', ds=ds)
for res, res_zv in ((res1, res1_zv), (res2, res2_zv)):
for f, f_zv in zip(res.f, res_zv.f):
assert_array_equal((f_zv == 0).x, zero_var.x)
assert f_zv[zv_index] == 0
f_zv[zv_index] = f[zv_index]
assert_dataobj_equal(f_zv, f, decimal=decimal)
# nested random effect
res = testnd.anova('uts', 'A * B * nrm(A)', ds=ds, samples=10, tstart=.4)
assert res.match == 'nrm(A)'
assert [p.min() for p in res.p] == [0.0, 0.6, 0.9]
# unequal argument length
with pytest.raises(ValueError):
testnd.anova('uts', 'A[:-1]', ds=ds)
with pytest.raises(ValueError):
testnd.anova('uts[:-1]', 'A * B * nrm(A)', ds=ds)
def test_anova():
"Test testnd.anova()"
ds = datasets.get_uts(True, nrm=True)
testnd.anova('utsnd', 'A*B', ds=ds)
for samples in (0, 2):
logging.info("TEST: samples=%r" % samples)
testnd.anova('utsnd', 'A*B', ds=ds, samples=samples)
testnd.anova('utsnd', 'A*B', ds=ds, samples=samples, pmin=0.05)
res = testnd.anova('utsnd', 'A*B', ds=ds, samples=samples, tfce=True)
assert res._plot_model() == 'A%B'
asfmtext(res)
res = testnd.anova('utsnd',
'A*B*rm',
match=False,
ds=ds,
samples=0,
pmin=0.05)
assert repr(
res
) == "<anova 'utsnd', 'A*B*rm', match=False, samples=0, pmin=0.05, 'A': 17 clusters, 'B': 20 clusters, 'A x B': 22 clusters>"
assert res._plot_model() == 'A%B'
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=2, pmin=0.05)
assert res.match == 'rm'
assert repr(
res
) == "<anova 'utsnd', 'A*B*rm', match='rm', samples=2, pmin=0.05, 'A': 17 clusters, p < .001, 'B': 20 clusters, p < .001, 'A x B': 22 clusters, p < .001>"
assert res._plot_model() == 'A%B'
# persistence
string = pickle.dumps(res, protocol=pickle.HIGHEST_PROTOCOL)
res_ = pickle.loads(string)
assert repr(res_) == repr(res)
assert res_._plot_model() == 'A%B'
# threshold-free
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=10)
assert res.match == 'rm'
assert repr(
res
) == "<anova 'utsnd', 'A*B*rm', match='rm', samples=10, 'A': p < .001, 'B': p < .001, 'A x B': p < .001>"
assert 'A clusters' in res.clusters.info
assert 'B clusters' in res.clusters.info
assert 'A x B clusters' in res.clusters.info
# no clusters
res = testnd.anova('uts',
'B',
sub="A=='a1'",
ds=ds,
samples=5,
pmin=0.05,
mintime=0.02)
repr(res)
assert 'v' in res.clusters
assert 'p' in res.clusters
assert res._plot_model() == 'B'
# all effects with clusters
res = testnd.anova('uts',
'A*B*rm',
match=False,
ds=ds,
samples=5,
pmin=0.05,
tstart=0.1,
mintime=0.02)
assert set(res.clusters['effect'].cells) == set(res.effects)
# some effects with clusters, some without
res = testnd.anova('uts',
'A*B*rm',
ds=ds,
samples=5,
pmin=0.05,
tstart=0.37,
mintime=0.02)
assert res.match == 'rm'
string = pickle.dumps(res, pickle.HIGHEST_PROTOCOL)
res_ = pickle.loads(string)
assert_dataobj_equal(res.clusters, res_.clusters)
# test multi-effect results (with persistence)
# UTS
res = testnd.anova('uts', 'A*B*rm', ds=ds, samples=5)
assert res.match == 'rm'
repr(res)
string = pickle.dumps(res, pickle.HIGHEST_PROTOCOL)
resr = pickle.loads(string)
tf_clusters = resr.find_clusters(pmin=0.05)
peaks = resr.find_peaks()
assert_dataobj_equal(tf_clusters, res.find_clusters(pmin=0.05))
assert_dataobj_equal(peaks, res.find_peaks())
assert tf_clusters.eval("p.min()") == peaks.eval("p.min()")
unmasked = resr.f[0]
masked = resr.masked_parameter_map(effect=0, pmin=0.05)
assert_array_equal(masked.x <= unmasked.x, True)
# reproducibility
decimal = 12 if IS_WINDOWS else None # FIXME: why is Windows sometimes different???
res0 = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
assert_dataset_equal(res.clusters, res0.clusters, decimal=decimal)
configure(n_workers=0)
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=5)
assert_dataset_equal(res.clusters, res0.clusters, decimal=decimal)
configure(n_workers=True)
# permutation
eelbrain._stats.permutation._YIELD_ORIGINAL = 1
samples = 4
# raw
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, samples=samples)
for dist in res._cdist:
assert len(dist.dist) == samples
assert_array_equal(dist.dist, dist.parameter_map.abs().max())
# TFCE
res = testnd.anova('utsnd', 'A*B*rm', ds=ds, tfce=True, samples=samples)
for dist in res._cdist:
assert len(dist.dist) == samples
assert_array_equal(dist.dist, dist.tfce_map.abs().max())
# thresholded
res1 = testnd.anova('utsnd', 'A*B*rm', ds=ds, pmin=0.05, samples=samples)
clusters = res1.find_clusters()
for dist, effect in zip(res1._cdist, res1.effects):
effect_idx = clusters.eval("effect == %r" % effect)
vmax = clusters[effect_idx, 'v'].abs().max()
assert len(dist.dist) == samples
assert_array_equal(dist.dist, vmax)
eelbrain._stats.permutation._YIELD_ORIGINAL = 0
# 1d TFCE
configure(n_workers=0)
res = testnd.anova('utsnd.rms(time=(0.1, 0.3))',
'A*B*rm',
ds=ds,
tfce=True,
samples=samples)
configure(n_workers=True)
# zero variance
res2 = testnd.anova('utsnd', 'A', ds=ds)
ds['utsnd'].x[:, 1, 10] = 0.
zero_var = ds['utsnd'].var('case') == 0
zv_index = tuple(i[0] for i in zero_var.nonzero())
res1_zv = testnd.anova('utsnd', 'A*B*rm', ds=ds)
res2_zv = testnd.anova('utsnd', 'A', ds=ds)
for res, res_zv in ((res1, res1_zv), (res2, res2_zv)):
for f, f_zv in zip(res.f, res_zv.f):
assert_array_equal((f_zv == 0).x, zero_var.x)
assert f_zv[zv_index] == 0
f_zv[zv_index] = f[zv_index]
assert_dataobj_equal(f_zv, f, decimal=decimal)
# nested random effect
res = testnd.anova('uts', 'A * B * nrm(A)', ds=ds, samples=10, tstart=.4)
assert res.match == 'nrm(A)'
assert [p.min() for p in res.p] == [0.0, 0.6, 0.9]
# unequal argument length
with pytest.raises(ValueError):
testnd.anova('uts', 'A[:-1]', ds=ds)
with pytest.raises(ValueError):
testnd.anova('uts[:-1]', 'A * B * nrm(A)', ds=ds)
def test_ttest():
"""Test univariate t-test functions"""
ds = datasets.get_uv()
print(test.ttest('fltvar', ds=ds))
print(test.ttest('fltvar', 'A', ds=ds))
print(test.ttest('fltvar', 'A%B', ds=ds))
print(test.ttest('fltvar', 'A', match='rm', ds=ds))
print(test.ttest('fltvar', 'A', 'a1', match='rm', ds=ds))
print(test.ttest('fltvar', 'A%B', ('a1', 'b1'), match='rm', ds=ds))
# Prepare data for scipy
a1_index = ds.eval("A == 'a1'")
a2_index = ds.eval("A == 'a2'")
b1_index = ds.eval("B == 'b1'")
a1_in_b1_index = np.logical_and(a1_index, b1_index)
a2_in_b1_index = np.logical_and(a2_index, b1_index)
# TTest1Samp
res = test.TTest1Sample('fltvar', ds=ds)
t, p = scipy.stats.ttest_1samp(ds['fltvar'], 0)
assert res.t == pytest.approx(t, 10)
assert res.p == pytest.approx(p, 10)
res = test.TTest1Sample('fltvar', ds=ds, tail=1)
assert res.t == pytest.approx(t, 10)
assert res.p == pytest.approx(p / 2., 10)
# TTestInd
res = test.TTestInd('fltvar', 'A', 'a1', 'a2', ds=ds)
t, p = scipy.stats.ttest_ind(ds[a1_index, 'fltvar'], ds[a2_index,
'fltvar'])
assert res.t == pytest.approx(t, 10)
assert res.p == pytest.approx(p, 10)
# TTestRel
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds)
a1 = ds[a1_in_b1_index, 'fltvar'].x
a2 = ds[a2_in_b1_index, 'fltvar'].x
difference = a1 - a2
t, p = scipy.stats.ttest_rel(a1, a2)
assert_array_equal(res.difference.x, difference)
assert res.df == len(a1) - 1
assert res.tail == 0
assert res.t == pytest.approx(t)
assert res.p == pytest.approx(p)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, 1)
assert_array_equal(res.difference.x, difference)
assert res.df == len(a1) - 1
assert res.tail == 1
assert res.t == pytest.approx(t)
assert res.p == pytest.approx(p / 2)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a2', 'a1', 'rm', "B=='b1'", ds, 1)
assert_array_equal(res.difference.x, -difference)
assert res.df == len(a1) - 1
assert res.tail == 1
assert res.t == pytest.approx(-t)
assert res.p == pytest.approx(1 - p / 2)
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, -1)
assert_array_equal(res.difference.x, difference)
assert res.df == len(a1) - 1
assert res.tail == -1
assert res.t == pytest.approx(t)
assert res.p == pytest.approx(1 - p / 2)
print(res)
print(asfmtext(res))
# alternative argspec
a1 = ds.eval("fltvar[(B == 'b1') & (A == 'a1')]")
a2 = ds.eval("fltvar[(B == 'b1') & (A == 'a2')]")
res_alt = test.TTestRel(a1, a2, tail=-1)
print(res_alt)
assert res_alt.t == res.t
assert res_alt.p == res.p
def test_ttest():
"""Test univariate t-test functions"""
ds = datasets.get_uv()
print(test.ttest('fltvar', ds=ds))
print(test.ttest('fltvar', 'A', ds=ds))
print(test.ttest('fltvar', 'A%B', ds=ds))
print(test.ttest('fltvar', 'A', match='rm', ds=ds))
print(test.ttest('fltvar', 'A', 'a1', match='rm', ds=ds))
print(test.ttest('fltvar', 'A%B', ('a1', 'b1'), match='rm', ds=ds))
# Prepare data for scipy
a1_index = ds.eval("A == 'a1'")
a2_index = ds.eval("A == 'a2'")
b1_index = ds.eval("B == 'b1'")
a1_in_b1_index = np.logical_and(a1_index, b1_index)
a2_in_b1_index = np.logical_and(a2_index, b1_index)
# TTest1Samp
res = test.TTest1Sample('fltvar', ds=ds)
t, p = scipy.stats.ttest_1samp(ds['fltvar'], 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
res = test.TTest1Sample('fltvar', ds=ds, tail=1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2., 10)
# TTestInd
res = test.TTestInd('fltvar', 'A', 'a1', 'a2', ds=ds)
t, p = scipy.stats.ttest_ind(ds[a1_index, 'fltvar'], ds[a2_index,
'fltvar'])
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
# TTestRel
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds)
a1 = ds[a1_in_b1_index, 'fltvar'].x
a2 = ds[a2_in_b1_index, 'fltvar'].x
diff = a1 - a2
t, p = scipy.stats.ttest_rel(a1, a2)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 0)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, 1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, 1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t > 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))
res = test.TTestRel('fltvar', 'A', 'a1', 'a2', 'rm', "B=='b1'", ds, -1)
assert_array_equal(res.diff.x, diff)
eq_(res.df, len(a1) - 1)
eq_(res.tail, -1)
assert_almost_equal(res.t, t, 10)
assert_almost_equal(res.p, p / 2 if t < 0 else 1 - p / 2, 10)
print(res)
print(asfmtext(res))