def run_emg_peak_fit(self,
t_peak_start: float,
t_peak_max: float,
t_peak_end: float,
relative_threshold: float):
rt_mean, sigma, skew = calc_emg_parameters_from_peak_shape(
t_peak_start, t_peak_max, t_peak_end, relative_threshold
)
v = peak_shapes.emg(np.array([t_peak_start, t_peak_max, t_peak_end]),
rt_mean, sigma, skew)
# try:
self.assertAlmostEqual(v[0], relative_threshold * v[1], 4)
self.assertAlmostEqual(v[2], relative_threshold * v[1], 4)
# except AssertionError as ex:
# print(ex)
# t = np.linspace(0, t_peak_end * 5, 10000)
# p = peak_shapes.emg(t, rt_mean, sigma, skew)
# from bokeh.plotting import figure, show
# plt_1 = figure(plot_width=690, plot_height=350)
# plt_1.line(t, p, line_width=2, color='green')
# plt_1.line(t_peak_start * np.ones(2),
# np.array([0, p.max()]),
# line_width=2, color='black')
# plt_1.line(t_peak_max * np.ones(2),
# np.array([0, p.max()]),
# line_width=2, color='green')
# plt_1.line(t_peak_end * np.ones(2),
# np.array([0, p.max()]),
# line_width=2, color='green')
# show(plt_1)
t = np.linspace(0, t_peak_end * 5, 10000)
p = peak_shapes.emg(t, rt_mean, sigma, skew)
self.assertAlmostEqual(p.max(), v[1], 3)
def setUp(self) -> None:
self.t = np.linspace(0, 100, 1500)
self.dt = self.t[1] - self.t[0]
self.log = TestLogger()
self.rtd_rt_mean = 50
self.rtd = peak_shapes.emg(self.t,
rt_mean=self.rtd_rt_mean,
sigma=20,
skew=0.5)
self.rtd_norm = self.rtd / self.rtd.sum() / self.dt
self.assertAlmostEqual(self.rtd_norm.sum() * self.dt, 1)
# define pre-fill
self.pre_fill = np.ones([3, 1])
self.pre_fill[1] = 0
self.pre_fill[2] = 0.23
# define wash
self.wash = np.ones([3, 1])
self.wash[0] = 0.45
self.wash[1] = 0.32
# define in c profile
self.c_in = np.ones([3, self.t.size])
self.c_in[0, :] = 0.1
self.c_in[1, :] = 0.5
self.c_in[:, -self.rtd.
size:] = 0 # clip the end in order to prevent the spill
def get_emg_pdf_with_cutoff(t, rt_mean, sigma, skew, cutoff) -> np.ndarray:
p: np.ndarray = peak_shapes.emg(t, rt_mean, sigma, skew)
# cut at end
p = p[:true_end(p >= cutoff * p.max())]
# set to 0 at front
p[:true_start(p >= cutoff * p.max())] = 0
# return normalized profile (integral = 1)
return p / p.sum() / t[1]
def score_func(x):
# x == [sigma, skew]
_sigma, _skew = x
_rt_mean = get_rt_mean(_sigma, _skew)
_p = _peak_shapes.emg(_t, _rt_mean, _sigma, _skew)
scr = (relative_threshold - _p[0] / _p[1]) ** 2 * 10 + \
(relative_threshold - _p[2] / _p[1]) ** 2
# print(_rt_mean, _sigma, _skew, scr)
return scr
def _calc_pdf(self, kw_pars: dict) -> _np.ndarray:
# Get `rt_mean` from parameters.
rt_mean = kw_pars['rt_mean'] if 'rt_mean' in kw_pars.keys() \
else kw_pars['v_void'] / kw_pars['f']
# Get `skew` from parameters if provided.
skew = kw_pars['skew'] if 'skew' in kw_pars.keys() else self.skew
# Calc `sigma`.
sigma = (rt_mean * self.dispersion_index)**0.5
# Calc probability distribution (`p`).
t = _np.arange(0, self._t_steps_max * self._dt, self._dt)
p = _peak_shapes.emg(t, rt_mean, sigma, skew, self.log)
return p
from bio_rtd import pdf, peak_shapes, peak_fitting
t = np.linspace(0, 10, 201)
# Calc rt_mean, sigma and skew from peak points.
rt, sigma, skew = peak_fitting.calc_emg_parameters_from_peak_shape(
t_peak_start=0.6, t_peak_max=1.4, t_peak_end=3.9, relative_threshold=0.1)
# Define pdf.
pdf_emg = pdf.ExpModGaussianFixedDispersion(t, sigma**2 / rt, skew)
pdf_emg.update_pdf(rt_mean=rt)
p_emg = pdf_emg.get_p()
# Generate noisy data.
y = peak_shapes.emg(t, 2, 0.3, 1.0) # clean signal
y_noisy = y + (np.random.random(y.shape) - 0.5) * y.max() / 10
# Plot.
p = figure(plot_width=690,
plot_height=350,
title="Probability Distribution",
x_axis_label="t [min]",
y_axis_label="c [mg/mL]")
p.line(t,
y_noisy,
line_width=2,
color='green',
alpha=0.6,
legend_label='c [mg/mL] (data)')
p.line(t, p_emg, line_width=2, color='black', alpha=1, legend_label='p (pdf)')
def test_emg(self) -> None:
t = np.linspace(0, 100, 1000)
rt_mean = 20
p = peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=2,
skew=1,
logger=self.log)
# validations
self.validate_rt_mean(t, p, rt_mean)
self.validate_point_on_slope(t, p, 15.33, 0.1 * p.max())
self.validate_point_on_slope(t, p, 24.80, 0.1 * p.max())
self.validate_point_on_slope(t, p, 17.35, 0.5 * p.max())
self.validate_point_on_slope(t, p, 22.47, 0.5 * p.max())
self.validate_max(t, p, 19.82)
# warnings
with self.assertWarns(Warning,
msg="EMG peak: sigma + 1 / skew < 4 * dt"):
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=4 * t[1] - 0.1 - 0.01,
skew=10,
logger=self.log)
with self.assertRaises(RuntimeError,
msg="Peak shape: integral: 0.7723355216030109"):
with self.assertWarns(Warning,
msg="EMG peak: rt_mean + 3 * sigma > t[-1]"):
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=(t[-1] - rt_mean) / 3 + 0.01,
skew=1,
logger=self.log)
with self.assertWarns(
Warning,
msg="EMG peak: t0 < 3 * sigma; t0 = rt_mean - 1 / skew"):
skew = 0.8
t0 = rt_mean - 1 / skew
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=t0 / 3 - 0.01,
skew=1,
logger=self.log)
with self.assertRaises(RuntimeError,
msg="Peak shape: integral: 0.5483504934228841"):
with self.assertWarns(Warning, msg="EMG peak: skew < 1/40"):
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=2,
skew=1 / 41,
logger=self.log)
with self.assertWarns(Warning, msg="EMG peak: skew > 10"):
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=2,
skew=10.1,
logger=self.log)
with self.assertRaises(
RuntimeError,
msg="Peak shape: relative value at start: 0.1359995541938657"):
with self.assertWarns(Warning,
msg="EMG peak: exp argument (p) > 200"):
skew = 2
sigma = 10
t0 = rt_mean - 1 / skew
assert np.any(skew / 2 * (2 *
(t0 - t) + skew * sigma**2) > 200)
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=sigma,
skew=skew,
logger=self.log)
with self.assertRaises(
RuntimeError,
msg="Peak shape: integral: 0.0069704155745190025"):
with self.assertWarns(Warning,
msg="EMG peak: exp argument (p) < -200"):
skew = 40
sigma = 1
t0 = rt_mean - 1 / skew
assert np.any(skew / 2 * (2 *
(t0 - t) + skew * sigma**2) < -200)
peak_shapes.emg(t,
rt_mean=rt_mean,
sigma=sigma,
skew=skew,
logger=self.log)
# 2nd test
t = np.linspace(0, 150, 5000)
rt_mean = 40
p = peak_shapes.emg(t, rt_mean, 7.4, 0.26, self.log)
# validations
self.validate_rt_mean(t, p, rt_mean)
self.validate_point_on_slope(t, p, 22.62725, 0.1 * p.max())
self.validate_point_on_slope(t, p, 57.90722, 0.1 * p.max())
self.validate_point_on_slope(t, p, 30.12018, 0.5 * p.max())
self.validate_point_on_slope(t, p, 49.14717, 0.5 * p.max())
self.validate_max(t, p, 39.41918)
