def test_unique_order():
for x, y in [(np.array([1, 2, 3, 4, 5, 6]), np.array([1, 3, -3, -7, -8])),
(rand(12), rand(222))]:
d1 = mpl.Data2D(x=x, y=y)
d2 = mpl.Data2D(xx=d1.xx, yy=d1.yy)
# test Data2D._unique()
#
# with np.unique()
# d1.y = array([1, 3, -3, -7, -8])
# d2.y = np.unique(d1.yy)
# = array([-8, -7, -3, 1, 3])
# instead of
# d2.y = Data2D._unique(d1.yy)
# = array([1, 3, -3, -7, -8])
# = d1.y
assert_equal(d1, d2)
def get_data():
x = np.linspace(-5, 5, 20)
y = x
X, Y = np.meshgrid(x, y)
X = X.T
Y = Y.T
Z = (X + 3)**2 + (Y + 4)**2 + 5
dd = mpl.Data2D(X=X, Y=Y, Z=Z)
tgt = np.array([5.0, 30])
return dd, tgt
def test_copy():
d1 = mpl.Data2D(x=rand(10), y=rand(12), zz=rand(10 * 12))
d2 = d1.copy()
assert_equal(d1, d2)
assert d1.zz is not None
assert d2.zz is not None
assert d1.Z is not None
assert d2.Z is not None
# zz and Z is None
d1 = mpl.Data2D(x=rand(10), y=rand(12))
d2 = d1.copy()
assert_equal(d1, d2)
assert d1.zz is None
assert d2.zz is None
assert d1.Z is None
assert d2.Z is None
# arrays are not views, changing d2.<attr> will not alter d1.<attr>
d2.xx *= 2.0
assert (d1.xx * 2.0 == d2.xx).all()
def test_update():
x1 = rand(10)
y1 = rand(12)
d1 = mpl.Data2D(x=x1, y=y1)
assert d1.zz is None
assert d1.Z is None
d1.update(Z=rand(10, 12))
assert d1.zz is not None
assert d1.Z is not None
d2 = d1.copy()
d2.update(x=rand(10), y=rand(12))
assert (d1.xx != d2.xx).all()
assert (d1.yy != d2.yy).all()
assert (d1.X != d2.X).all()
assert (d1.Y != d2.Y).all()
assert (d1.Z == d2.Z).all()
assert (d1.zz == d2.zz).all()
def test_input():
d1 = mpl.get_2d_testdata()
for k in [d1.x, d1.y, d1.xx, d1.yy, d1.zz, d1.X, d1.Y, d1.Z, d1.XY]:
assert k is not None
XY = np.array([k for k in itertools.product(d1.x, d1.y)])
zz = np.empty(np.prod(d1.Z.shape))
nx = d1.Z.shape[0]
ny = d1.Z.shape[1]
for ix in range(nx):
zz[ix * ny:(ix + 1) * ny] = d1.Z[ix, :]
xx = XY[:, 0]
yy = XY[:, 1]
assert nx == d1.nx
assert ny == d1.ny
assert (d1.xx == xx).all()
assert (d1.yy == yy).all()
assert (d1.XY == XY).all()
assert (d1.zz == d1.Z.flatten()).all()
assert (d1.zz == zz).all()
# test various forms of x-y input
d2 = mpl.Data2D(x=d1.x, y=d1.y, zz=d1.zz)
assert_equal(d1, d2)
d2 = mpl.Data2D(x=d1.x, y=d1.y, Z=d1.Z)
assert_equal(d1, d2)
d2 = mpl.Data2D(xx=d1.xx, yy=d1.yy, Z=d1.Z)
assert_equal(d1, d2)
d2 = mpl.Data2D(X=d1.X, Y=d1.Y, Z=d1.Z)
assert_equal(d1, d2)
d2 = mpl.Data2D(XY=d1.XY, Z=d1.Z)
assert_equal(d1, d2)
# Z data is optional
d2 = mpl.Data2D(XY=d1.XY)
assert_equal(d1, d2, keys=['x', 'y', 'xx', 'yy', 'X', 'Y', 'XY'])
rnd = np.random.RandomState(seed=1234)
for name in ['gauss', 'multi', 'inv_multi']:
print(name)
fig,ax = mpl.fig_ax()
ax.set_title(name)
x = np.linspace(0, 10, npoints)
y = np.sin(x) + rnd.rand(len(x))
fe = rbf.FitError(x[:,None], y, cv_kwds=dict(ns=5, nr=1),
rbf_kwds=dict(rbf=name))
p = np.linspace(0.01, 15, nsample)
r = np.logspace(-10, 2, nsample)
g = np.array(list(itertools.product(p,r)))
zz = np.array([np.median(fe.err_cv(params)) for params in g])
zz[zz > zmax] = zmax
dd = mpl.Data2D(xx=g[:,0], yy=np.log10(g[:,1]), zz=zz)
pl = ax.contourf(dd.X, dd.Y, dd.Z, cmap=cm.jet)
plt.colorbar(pl)
ax.set_xlabel('p')
ax.set_ylabel('log10(r)')
## fig.savefig(f'/tmp/crossval_pr_{name}.png')
## f = rbf.fit_opt(x[:,None], y, method='de', what='pr',
## opt_kwds=dict(disp=True,
## maxiter=20,
## popsize=20,
## bounds=[(p.min(),p.max()),(r.min(),r.max())],
## mutation=1.5,
## polish=True),
## rbf_kwds=dict(rbf=name))
## popt = f.get_params()