def cumplot(ax, x, y, label):
x = uf.cumsum(x)
y = uf.cumsum(y)
mask = (~uf.isnan(x) & ~uf.isnan(y) & ~uf.isinf(x) & ~uf.isinf(y))
line, = ax.plot(x.values[mask.values], y.values[mask.values], label=label)
ax.set_ylabel(y.attrs['quantity'] + ' (' + y.attrs['unit'] + ')')
ax.set_xlabel(x.attrs['quantity'] + ' (' + x.attrs['unit'] + ')')
def interm(x, y, axis=None):
n = np.nansum((x > 0.1) & (y > 0.1) & ~ufuncs.isnan(x) & ~ufuncs.isnan(y),
axis=axis)
o = np.nansum(
((x > 0.1) | (y > 0.1)) & ~ufuncs.isnan(x) & ~ufuncs.isnan(y),
axis=axis)
return n / o
def timeseries(array):
fig, ax = pl.subplots()
array = array[~uf.isnan(array)]
line, = ax.plot(array['time'].values, array.values, label=array.name)
ax.set_ylabel(array.attrs['quantity'] + ' (' + array.attrs['unit'] + ')')
ax.legend()
ax.set_xlabel('time')
pl.grid = True
return fig, ax
def kde(x, y, bandwidth=1.0, log=False, num=200, **kwargs):
"""
kernel density estimation
"""
typ = type(x)
xdim = x.dims
ydim = y.dims
if xdim != ydim:
err = 'unequal dimensions: %s and %s' % (str(xdim), str(ydim))
raise Exception(err)
# prepare training data
if log is True:
x, y = uf.log(x), uf.log(y)
x_m, y_m = uf.nanmean(x), uf.nanmean(y)
x_s, y_s = uf.nanstd(x), uf.nanstd(y)
x = (x - x_m) / x_s
y = (y - y_m) / y_s
# estimate bandwidth with cross validation
mask = (~uf.isnan(y) & ~uf.isnan(x) & ~uf.isinf(y) & ~uf.isinf(x))
train = np.vstack((x.values[mask.values], y.values[mask.values])).T
if bandwidth is None:
gcv = GridSearchCV(KernelDensity(),
{'bandwidth': np.linspace(0.05, 1.0)},
cv=5)
gcv.fit(train)
bandwidth = gcv.best_params_['bandwidth']
kde = KernelDensity(bandwidth=bandwidth).fit(train)
# prepare sample grid
x_g = np.linspace(uf.nanmin(x).values, uf.nanmax(x).values, num=num)
y_g = np.linspace(uf.nanmin(y).values, uf.nanmax(y).values, num=num)
x_grid, y_grid = np.meshgrid(x_g, y_g)
grid = np.vstack((x_grid.flatten(), y_grid.flatten())).T
z = np.exp(kde.score_samples(grid)).reshape(x_grid.shape).T
x = Index(x_g * x_s + x_m, name=x.name, attrs=x.attrs)
y = Index(y_g * y_s + y_m, name=y.name, attrs=y.attrs)
return typ(z,
coords=[x, y],
name='density',
attrs={'bandwidth': bandwidth})
def linregress2(x, y, prob=0.95, **kwargs):
"""
linear least-squares regression
"""
c = uf.cov(x, y).task
r = uf.corr(x, y).task[0, 1]
a1 = (c[0, 1] / c[0, 0])
cond = uf.isnan(x) | uf.isnan(y)
x = uf.where(cond, np.nan, x)
y = uf.where(cond, np.nan, y)
n = uf.nansum(~uf.isnan(x)).task
a0 = uf.nanmean(y) - uf.nanmean(x) * a1
stderr = uf.sqrt((1 - r**2) * c[1, 1] / c[0, 0] / (n - 2))
stderr_res = uf.nanstd(y - a0 - x * a1) * ((n - 1) / (n - 2))**0.5
stderr_a1 = stderr_res / uf.nanstd(x) * n**-0.5
return da.compute(a0.task, a1, r, stderr, stderr_res.task, stderr_a1.task,
c, n)