def test_variability():
"Test variability functions"
ds = datasets.get_loftus_masson_1994()
y = ds['n_recalled'].x.astype(np.float64)
x = ds['exposure'].as_factor()
match = ds['subject']
sem = scipy.stats.sem(y, 0, 1)
ci = sem * scipy.stats.t.isf(0.05 / 2., len(y) - 1)
# invalid spec
assert_raises(ValueError, stats.variability, y, 0, 0, '1mile', 0)
assert_raises(ValueError, stats.variability, y, 0, 0, 'ci7ci', 0)
# standard error
assert_almost_equal(stats.variability(y, None, None, 'sem', False), sem)
assert_almost_equal(stats.variability(y, None, None, '2sem', False), 2 * sem)
# within subject standard-error
target = scipy.stats.sem(stats.residuals(y[:, None], match), 0, len(match.cells))
assert_almost_equal(stats.variability(y, None, match, 'sem', True), target)
assert_almost_equal(stats.variability(y, None, match, 'sem', False), target)
# one data point per match cell
n = match.df + 1
assert_raises(ValueError, stats.variability, y[:n], None, match[:n], 'sem', True)
target = np.array([scipy.stats.sem(y[x == cell], 0, 1) for cell in x.cells])
es = stats.variability(y, x, None, 'sem', False)
assert_allclose(es, target)
stats.variability(y, x, None, 'sem', True)
# confidence intervals
assert_almost_equal(stats.variability(y, None, None, '95%ci', False), ci)
assert_almost_equal(stats.variability(y, x, None, '95%ci', True), 3.86, 2) # L&M: 3.85
assert_almost_equal(stats.variability(y, x, match, '95%ci', True), 0.52, 2)
assert_equal(stats.variability(y, x, None, '95%ci', False)[::-1],
stats.variability(y, x, None, '95%ci', False, x.cells[::-1]))
def _plt_barplot(ax, ct, error, pool_error, hatch, colors, bottom, top=None,
origin=None, left=None, width=.5, c='#0099FF', edgec=None,
ec='k', test=True, par=True, trend="'", corr='Hochberg',
test_markers=True):
"""Draw a barplot to axes ax for Celltable ct.
Parameters
----------
ax : mpl Axes
Axes to which to plot
ct : Celltable
Data to plot.
error : str
Variability description (e.g., "95%ci").
pool_error : bool
Pool the errors for the estimate of variability.
...
"""
# kwargs
if hatch is True:
hatch = defaults['hatch']
if colors is True:
if defaults['mono']:
colors = defaults['cm']['colors']
else:
colors = defaults['c']['colors']
elif isinstance(colors, dict):
colors = [colors[cell] for cell in ct.cells]
# data means
k = len(ct.cells)
if left is None:
left = np.arange(k) - width / 2
height = np.array(ct.get_statistic(np.mean))
# origin
if origin is None:
origin = max(0, bottom)
# error bars
if ct.X is None:
error_match = None
else:
error_match = ct.match
y_error = stats.variability(ct.Y.x, ct.X, error_match, error, pool_error, ct.cells)
# fig spacing
plot_max = np.max(height + y_error)
plot_min = np.min(height - y_error)
plot_span = plot_max - plot_min
y_bottom = min(bottom, plot_min - plot_span * .05)
# main BARPLOT
bars = ax.bar(left, height - origin, width, bottom=origin, align='edge',
color=c, edgecolor=edgec, ecolor=ec, yerr=y_error)
# hatch
if hatch:
for bar, h in zip(bars, hatch):
bar.set_hatch(h)
if colors:
for bar, c in zip(bars, colors):
bar.set_facecolor(c)
# pairwise tests
if ct.X is None and test is True:
test = 0.
y_unit = (plot_max - y_bottom) / 15
if test is True:
y_top = _mark_plot_pairwise(ax, ct, par, plot_max, y_unit, corr, trend,
test_markers, top=top)
elif (test is False) or (test is None):
y_top = plot_max + y_unit
else:
ax.axhline(test, color='black')
y_top = _mark_plot_1sample(ax, ct, par, plot_max, y_unit, test, corr, trend)
if top is None:
top = y_top
# x0, x1, y0, y1
lim = (min(left) - .5 * width, max(left) + 1.5 * width, y_bottom, top)
return lim
def test_sem_and_variability():
"Test variability() and standard_error_of_the_mean() functions"
ds = datasets.get_loftus_masson_1994()
y = ds['n_recalled'].x
x = ds['exposure'].as_factor()
match = ds['subject']
# invalid spec
assert_raises(ValueError, stats.variability, y, 0, 0, '1mile', 0)
assert_raises(ValueError, stats.variability, y, 0, 0, 'ci7ci', 0)
# standard error
target = scipy.stats.sem(y, 0, 1)
e = stats.variability(y, None, None, 'sem', False)
assert_almost_equal(e, target)
e = stats.variability(y, None, None, '2sem', False)
assert_almost_equal(e, 2 * target)
# within subject standard-error
target = scipy.stats.sem(stats.residuals(y[:, None], match), 0, len(match.cells))
assert_almost_equal(stats.variability(y, None, match, 'sem', True), target)
assert_almost_equal(stats.variability(y, None, match, 'sem', False), target)
# one data point per match cell
n = match.df + 1
assert_raises(ValueError, stats.variability, y[:n], None, match[:n], 'sem', True)
target = np.array([scipy.stats.sem(y[x == cell], 0, 1) for cell in x.cells])
es = stats.variability(y, x, None, 'sem', False)
assert_allclose(es, target)
stats.variability(y, x, None, 'sem', True)
# confidence intervals
stats.variability(y, None, None, '95%ci', False)
stats.variability(y, x, None, '95%ci', True)
stats.variability(y, x, match, '95%ci', True)
def test_variability():
"Test variability functions"
ds = datasets.get_loftus_masson_1994()
y = ds['n_recalled'].x.astype(np.float64)
x = ds['exposure'].as_factor()
match = ds['subject']
sem = scipy.stats.sem(y, 0, 1)
ci = sem * scipy.stats.t.isf(0.05 / 2., len(y) - 1)
# invalid spec
with pytest.raises(ValueError):
stats.variability(y, 0, 0, '1mile', 0)
with pytest.raises(ValueError):
stats.variability(y, 0, 0, 'ci7ci', 0)
# standard error
assert stats.variability(y, None, None, 'sem', False) == sem
assert stats.variability(y, None, None, '2sem', False) == 2 * sem
# within subject standard-error
target = scipy.stats.sem(stats.residuals(y[:, None], match), 0,
len(match.cells))[0]
assert stats.variability(y, None, match, 'sem',
True) == pytest.approx(target)
assert stats.variability(y, None, match, 'sem',
False) == pytest.approx(target)
# one data point per match cell
n = match.df + 1
with pytest.raises(ValueError):
stats.variability(y[:n], None, match[:n], 'sem', True)
target = np.array(
[scipy.stats.sem(y[x == cell], 0, 1) for cell in x.cells])
es = stats.variability(y, x, None, 'sem', False)
assert_allclose(es, target)
stats.variability(y, x, None, 'sem', True)
# confidence intervals
assert stats.variability(y, None, None, '95%ci',
False) == pytest.approx(ci)
assert stats.variability(y, x, None, '95%ci',
True) == pytest.approx(3.86,
abs=1e-2) # L&M: 3.85
assert stats.variability(y, x, match, '95%ci',
True) == pytest.approx(0.52, abs=1e-2)
assert_equal(
stats.variability(y, x, None, '95%ci', False)[::-1],
stats.variability(y, x, None, '95%ci', False, x.cells[::-1]))
def _plt_barplot(ax, ct, error, pool_error, hatch, colors, bottom, top=None,
origin=None, left=None, width=.5, c='#0099FF', edgec=None,
ec='k', test=True, par=True, trend="'", corr='Hochberg',
test_markers=True):
"""Draw a barplot to axes ax for Celltable ct.
Parameters
----------
ax : mpl Axes
Axes to which to plot
ct : Celltable
Data to plot.
error : str
Variability description (e.g., "95%ci").
pool_error : bool
Pool the errors for the estimate of variability.
...
"""
# kwargs
if hatch is True:
hatch = defaults['hatch']
if colors is True:
if defaults['mono']:
colors = defaults['cm']['colors']
else:
colors = defaults['c']['colors']
elif isinstance(colors, dict):
colors = [colors[cell] for cell in ct.cells]
# data means
k = len(ct.cells)
if left is None:
left = np.arange(k) - width / 2
height = np.array(ct.get_statistic(np.mean))
# origin
if origin is None:
origin = max(0, bottom)
# error bars
if ct.X is None:
error_match = None
else:
error_match = ct.match
y_error = stats.variability(ct.Y.x, ct.X, error_match, error, pool_error)
# fig spacing
plot_max = np.max(height + y_error)
plot_min = np.min(height - y_error)
plot_span = plot_max - plot_min
y_bottom = min(bottom, plot_min - plot_span * .05)
# main BARPLOT
bars = ax.bar(left, height - origin, width, bottom=origin,
color=c, edgecolor=edgec, ecolor=ec, yerr=y_error)
# hatch
if hatch:
for bar, h in zip(bars, hatch):
bar.set_hatch(h)
if colors:
for bar, c in zip(bars, colors):
bar.set_facecolor(c)
# pairwise tests
if ct.X is None and test is True:
test = 0.
y_unit = (plot_max - y_bottom) / 15
if test is True:
y_top = _mark_plot_pairwise(ax, ct, par, plot_max, y_unit, corr, trend,
test_markers, top=top)
elif (test is False) or (test is None):
y_top = plot_max + y_unit
else:
ax.axhline(test, color='black')
y_top = _mark_plot_1sample(ax, ct, par, plot_max, y_unit, test, corr, trend)
if top is None:
top = y_top
# x0, x1, y0, y1
lim = (min(left) - .5 * width, max(left) + 1.5 * width, y_bottom, top)
return lim
def test_sem_and_variability():
"Test variability() and standard_error_of_the_mean() functions"
ds = datasets.get_loftus_masson_1994()
y = ds['n_recalled'].x.astype(np.float64)
x = ds['exposure'].as_factor()
match = ds['subject']
# invalid spec
assert_raises(ValueError, stats.variability, y, 0, 0, '1mile', 0)
assert_raises(ValueError, stats.variability, y, 0, 0, 'ci7ci', 0)
# standard error
target = scipy.stats.sem(y, 0, 1)
e = stats.variability(y, None, None, 'sem', False)
assert_almost_equal(e, target)
e = stats.variability(y, None, None, '2sem', False)
assert_almost_equal(e, 2 * target)
# within subject standard-error
target = scipy.stats.sem(stats.residuals(y[:, None], match), 0,
len(match.cells))
assert_almost_equal(stats.variability(y, None, match, 'sem', True), target)
assert_almost_equal(stats.variability(y, None, match, 'sem', False),
target)
# one data point per match cell
n = match.df + 1
assert_raises(ValueError, stats.variability, y[:n], None, match[:n], 'sem',
True)
target = np.array(
[scipy.stats.sem(y[x == cell], 0, 1) for cell in x.cells])
es = stats.variability(y, x, None, 'sem', False)
assert_allclose(es, target)
stats.variability(y, x, None, 'sem', True)
# confidence intervals
stats.variability(y, None, None, '95%ci', False)
stats.variability(y, x, None, '95%ci', True)
stats.variability(y, x, match, '95%ci', True)
assert_equal(
stats.variability(y, x, None, '95%ci', False)[::-1],
stats.variability(y, x, None, '95%ci', False, x.cells[::-1]))
def __init__(self, ax, ct, error, pool_error, hatch, colors, bottom, top=None,
origin=None, left=None, width=.5, c='#0099FF', edgec=None,
ec='k', test=True, par=True, trend="'", corr='Hochberg',
test_markers=True, xticks=None, xtick_delim=None):
# kwargs
if hatch is True:
hatch = defaults['hatch']
if colors is True:
if defaults['mono']:
colors = defaults['cm']['colors']
else:
colors = defaults['c']['colors']
elif isinstance(colors, dict):
colors = [colors[cell] for cell in ct.cells]
# data means
k = len(ct.cells)
if left is None:
left = np.arange(k) - width / 2
height = np.array(ct.get_statistic(np.mean))
# origin
if origin is None:
origin = max(0, bottom)
# error bars
if ct.x is None:
error_match = None
else:
error_match = ct.match
y_error = stats.variability(ct.y.x, ct.x, error_match, error, pool_error, ct.cells)
# fig spacing
plot_max = np.max(height + y_error)
plot_min = np.min(height - y_error)
plot_span = plot_max - plot_min
y_bottom = min(bottom, plot_min - plot_span * .05)
# main BARPLOT
bars = ax.bar(left, height - origin, width, bottom=origin, align='edge',
color=c, edgecolor=edgec, ecolor=ec, yerr=y_error)
# hatch
if hatch:
for bar, h in zip(bars, hatch):
bar.set_hatch(h)
if colors:
for bar, c in zip(bars, colors):
bar.set_facecolor(c)
# pairwise tests
if ct.x is None and test is True:
test = 0.
y_unit = (plot_max - y_bottom) / 15
if test is True:
y_top = _mark_plot_pairwise(ax, ct, par, plot_max, y_unit, corr, trend,
test_markers, top=top)
elif (test is False) or (test is None):
y_top = plot_max + y_unit
else:
ax.axhline(test, color='black')
y_top = _mark_plot_1sample(ax, ct, par, plot_max, y_unit, test, corr, trend)
self.left = min(left) - .5 * width
self.right = max(left) + 1.5 * width
self.bottom = y_bottom
self.top = y_top if top is None else top
_plt_uv_base.__init__(self, ax, ct, xticks, xtick_delim)