Skip to content

Commit 24fac9d

Browse files
committed
Merge pull request #1106 from astrofrog/modeling/fix-gaussian-stddev
Gaussian1dModel.stddev is incorrectly used This also fixed points 1 and 2 in #1077
2 parents b5df9fb + 5de9a89 commit 24fac9d

4 files changed

Lines changed: 106 additions & 66 deletions

File tree

‎astropy/modeling/functional_models.py‎

Lines changed: 91 additions & 53 deletions
Original file line numberDiff line numberDiff line change
@@ -8,12 +8,13 @@
88
from . import parameters
99
from .core import *
1010
from .utils import InputParameterError
11-
11+
1212
__all__ = ['Gaussian1DModel', 'Gaussian2DModel', 'ScaleModel', 'ShiftModel']
13-
13+
14+
1415
class Gaussian1DModel(ParametricModel):
1516
"""
16-
17+
1718
Implements 1D Gaussian model.
1819
1920
Parameters
@@ -22,85 +23,97 @@ class Gaussian1DModel(ParametricModel):
2223
Amplitude of the gaussian
2324
mean : float
2425
Mean of the gaussian
25-
fwhm : float
26-
FWHM
2726
stddev : float
2827
Standard deviation of the gaussian
2928
Either fwhm or stddev must be specified
29+
fwhm : float
30+
Full width at half maximum
31+
Either fwhm or stddev must be specified
3032
jacobian_func : callable, 'estimated' or None
31-
if callable - a function to compute the Jacobian of
33+
if callable - a function to compute the Jacobian of
3234
func with derivatives across the rows.
3335
if 'estimated' - the Jacobian will be estimated
3436
if None - if the model has a deriv method, it will be used,
3537
if not the Jacobian will be estimated.
36-
38+
3739
"""
40+
3841
param_names = ['amplitude', 'mean', 'stddev']
39-
def __init__(self, amplitude, mean, fwhm=None, stddev=None,
42+
43+
def __init__(self, amplitude, mean, stddev=None, fwhm=None,
4044
jacobian_func=None, **cons):
45+
4146
try:
4247
param_dim = len(amplitude)
4348
except TypeError:
4449
param_dim = 1
50+
4551
if stddev is None and fwhm is None:
4652
raise InputParameterError(
4753
"Either fwhm or stddev must be specified")
48-
if stddev is not None:
54+
elif stddev is not None and fwhm is not None:
55+
raise InputParameterError("Cannot specify both fwhm and stddev")
56+
elif stddev is not None:
4957
stddev_val = stddev
5058
else:
5159
try:
5260
stddev_val = 0.42466 * fwhm
5361
except TypeError:
5462
stddev_val = [0.42466 * n for n in fwhm]
63+
5564
self._stddev = parameters.Parameter('stddev', stddev_val, self, param_dim)
5665
self._mean = parameters.Parameter('mean', mean, self, param_dim)
5766
self._amplitude = parameters.Parameter('amplitude', amplitude, self, param_dim)
67+
5868
super(Gaussian1DModel, self).__init__(self.param_names, n_inputs=1, n_outputs=1,
5969
param_dim=param_dim, **cons)
70+
6071
self.linear = False
6172
if jacobian_func is 'estimated':
6273
self.deriv = None
6374
elif callable(jacobian_func):
6475
self.deriv = jacobian_func
6576
else:
6677
self.deriv = self.gderiv
67-
78+
6879
def eval(self, x, params):
69-
return params[0] * np.exp((-(1/(params[2]**2)) *
70-
(x-params[1])**2))
71-
80+
return params[0] * np.exp(- 0.5 * (x-params[1])**2 / params[2]**2)
81+
7282
def gderiv(self, p, x, y):
7383
amplitude, mean, stddev = p
7484
deriv_dict = {}
75-
deriv_dict['amplitude'] = np.exp((-(1/(stddev**2)) * (x-mean)**2))
76-
deriv_dict['mean'] = 2 * amplitude * np.exp((-(1/(stddev**2)) *
77-
(x-mean)**2)) * (x-mean)/(stddev**2)
78-
deriv_dict['stddev'] = 2 * amplitude * np.exp((-(1/(stddev**2)) *
79-
(x-mean)**2)) * ((x-mean)**2)/(stddev**3)
85+
deriv_dict['amplitude'] = np.exp(-0.5 / stddev**2 * (x-mean)**2)
86+
deriv_dict['mean'] = (amplitude
87+
* np.exp(-0.5 / stddev**2 * (x-mean)**2)
88+
* (x-mean) / stddev**2)
89+
deriv_dict['stddev'] = (amplitude
90+
* np.exp(-0.5 / stddev**2 * (x-mean)**2)
91+
* (x-mean)**2 / stddev**3)
8092
derivval = [deriv_dict[par] for par in self.param_names]
8193
return np.array(derivval).T
82-
94+
8395
def __call__(self, x):
8496
"""
8597
Transforms data using this model.
86-
98+
8799
Parameters
88100
--------------
89101
x : array, of minimum dimensions 1
90-
102+
91103
Notes
92104
-----
93-
See the module docstring for rules for model evaluation.
105+
See the module docstring for rules for model evaluation.
94106
"""
95107
x, fmt = _convert_input(x, self.param_dim)
96108
result = self.eval(x, self.param_sets)
97109
return _convert_output(result, fmt)
98-
110+
111+
99112
class Gaussian2DModel(ParametricModel):
100113
"""
101-
114+
102115
2D Gaussian.
103-
116+
104117
Parameters
105118
----------
106119
amplitude : float
@@ -109,36 +122,41 @@ class Gaussian2DModel(ParametricModel):
109122
Mean of the gaussian in x
110123
y_mean : float
111124
Mean of the gaussian in y
112-
x_fwhm : float
113-
Full width at half maximum in x
114-
y_fwhm : float
115-
Full width at half maximum in y
116125
x_stddev : float
117126
Standard deviation of the gaussian in x
118-
Either fwhm or x_stddev must be specified
127+
Either x_fwhm or x_stddev must be specified
119128
y_stddev : float
120129
Standard deviation of the gaussian in y
121-
Either y_stddev or ratio should be given
130+
Either y_fwhm or y_stddev must be specified
131+
theta : float
132+
Rotation angle in radians. Note: increases clockwise.
133+
x_fwhm : float
134+
Full width at half maximum in x
135+
Either x_fwhm or x_stddev must be specified
136+
y_fwhm : float
137+
Full width at half maximum in y
138+
Either y_fwhm or y_stddev must be specified
122139
jacobian_func : callable or None
123-
if callable - a function to compute the Jacobian of
140+
if callable - a function to compute the Jacobian of
124141
func with derivatives across the rows.
125142
if None - the Jacobian will be estimated
126-
theta : float
127-
Rotation angle in radians. Note: increases clockwise.
128143
cov_matrix : ndarray
129-
A 2x2 covariance matrix. If specified, overrides stddev, fwhm, and
144+
A 2x2 covariance matrix. If specified, overrides stddev, fwhm, and
130145
theta specification.
131146
"""
132-
param_names = ['amplitude', 'x_mean', 'y_mean', \
147+
148+
param_names = ['amplitude', 'x_mean', 'y_mean',
133149
'x_stddev', 'y_stddev', 'theta']
134-
135-
def __init__(self, amplitude, x_mean, y_mean, x_fwhm=None, y_fwhm=None,
136-
x_stddev=None, y_stddev=None, theta=0.0, cov_matrix=None,
150+
151+
def __init__(self, amplitude, x_mean, y_mean, x_stddev=None, y_stddev=None,
152+
theta=0.0, x_fwhm=None, y_fwhm=None, cov_matrix=None,
137153
jacobian_func=None, **cons):
154+
138155
try:
139156
param_dim = len(amplitude)
140157
except TypeError:
141158
param_dim = 1
159+
142160
if y_stddev is None and y_fwhm is None and cov_matrix is None:
143161
raise InputParameterError(
144162
"Either y_fwhm or y_stddev must be specified, or a "
@@ -147,6 +165,16 @@ def __init__(self, amplitude, x_mean, y_mean, x_fwhm=None, y_fwhm=None,
147165
raise InputParameterError(
148166
"Either x_fwhm or x_stddev must be specified, or a "
149167
"covariance matrix.")
168+
elif x_stddev is not None and x_fwhm is not None:
169+
raise InputParameterError("Cannot specify both x_fwhm and x_stddev")
170+
elif y_stddev is not None and y_fwhm is not None:
171+
raise InputParameterError("Cannot specify both y_fwhm and y_stddev")
172+
elif cov_matrix is not None and (x_stddev is not None or
173+
y_stddev is not None or
174+
x_fwhm is not None or
175+
y_fwhm is not None):
176+
raise InputParameterError("Cannot specify both cov_matrix and x/y_stddev or x/y_fwhm")
177+
150178
self._amplitude = parameters.Parameter('amplitude', amplitude,
151179
self, param_dim)
152180
if cov_matrix is None:
@@ -161,13 +189,15 @@ def __init__(self, amplitude, x_mean, y_mean, x_fwhm=None, y_fwhm=None,
161189
x_stddev, y_stddev = np.sqrt(eig_vals)
162190
y_vec = eig_vecs[:,0]
163191
theta = np.arctan2(y_vec[1],y_vec[0])
192+
164193
self._x_stddev = parameters.Parameter('x_stddev', x_stddev,
165194
self, param_dim)
166195
self._y_stddev = parameters.Parameter('y_stddev', y_stddev,
167196
self, param_dim)
168197
self._x_mean = parameters.Parameter('x_mean', x_mean, self, param_dim)
169198
self._y_mean = parameters.Parameter('y_mean', y_mean, self, param_dim)
170199
self._theta = parameters.Parameter('theta', theta, self, param_dim)
200+
171201
super(Gaussian2DModel, self).__init__(self.param_names, n_inputs=2, n_outputs=1,
172202
param_dim=param_dim, **cons)
173203
self.linear = False
@@ -177,27 +207,34 @@ def __init__(self, amplitude, x_mean, y_mean, x_fwhm=None, y_fwhm=None,
177207
self.deriv = None
178208

179209
def eval(self, x, y, p):
180-
return p[0] * np.exp(-(
181-
((np.cos(p[5])/p[1])**2 + (np.sin(p[5])/p[2])**2) * ((x-p[3])**2)
182-
+ 2*(np.cos(p[5])*np.sin(p[5])*(1/(p[1]**2)-1/(p[2]**2))) *
183-
(x-p[3])*(y-p[4]) + ((np.sin(p[5])/p[1])**2+(np.cos(p[5])/p[2])**2)*
184-
((y-p[4])**2)))
185-
210+
211+
a = 0.5 * ((np.cos(p[5]) / p[3]) ** 2 +
212+
(np.sin(p[5]) / p[4]) ** 2)
213+
b = 0.5 * (np.cos(p[5]) * np.sin(p[5]) *
214+
(1. / p[3] ** 2 - 1. / p[4] ** 2))
215+
c = 0.5 * ((np.sin(p[5]) / p[3]) ** 2 +
216+
(np.cos(p[5]) / p[4]) ** 2)
217+
218+
return p[0] * np.exp(-(a * (x - p[1]) ** 2 +
219+
b * (x - p[1]) * (y - p[2]) +
220+
c * (y - p[2]) ** 2))
221+
186222
def __call__(self, x, y):
187223
"""
188224
Transforms data using this model.
189-
225+
190226
Parameters
191227
--------------
192228
x, y : arrays, of min dimensions 2
193-
194-
Note: See the module docstring for rules for model evaluation.
229+
230+
Note: See the module docstring for rules for model evaluation.
195231
"""
196232
x, _ = _convert_input(x, self.param_dim)
197233
y, fmt = _convert_input(y, self.param_dim)
198234
result = self.eval(x, y, self.param_sets)
199235
return _convert_output(result, fmt)
200236

237+
201238
class ShiftModel(Model):
202239
"""
203240
Shift a coordinate.
@@ -208,7 +245,7 @@ class ShiftModel(Model):
208245
offsets to be applied to a coordinate
209246
if a list - each value in the list is an offset to be applied to a
210247
column in the input coordinate array
211-
248+
212249
"""
213250
param_names = ['offsets']
214251
def __init__(self, offsets):
@@ -227,17 +264,18 @@ def __call__(self, x):
227264
x, fmt = _convert_input(x, self.param_dim)
228265
result = x + self.offsets
229266
return _convert_output(result, fmt)
230-
267+
268+
231269
class ScaleModel(Model):
232270
"""
233-
271+
234272
Multiply a model by a factor.
235273
236274
Parameters
237275
---------------
238276
factors : float or a list of floats
239277
scale for a coordinate
240-
278+
241279
"""
242280
param_names = ['factors']
243281
def __init__(self, factors):
@@ -248,7 +286,7 @@ def __init__(self, factors):
248286
self._factors = parameters.Parameter('factors', factors, self, param_dim)
249287
super(ScaleModel, self).__init__(self.param_names, n_inputs=1, n_outputs=1,
250288
param_dim=param_dim)
251-
289+
252290
def __call__(self, x):
253291
"""
254292
Transforms data using this model.

‎astropy/modeling/tests/test_constraints.py‎

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -60,8 +60,7 @@ def testJointFitter(self):
6060
p2 = [13, .4]
6161
A = 9.8
6262
p = np.r_[A, p1, p2]
63-
compmodel = lambda A, p, x: A* np.exp((-(1/(p[1]**2)) *
64-
(x-p[0])**2))
63+
compmodel = lambda A, p, x: A* np.exp(-0.5 / p[1]**2 * (x-p[0])**2)
6564
errf = lambda p, x1, y1, x2, y2: np.ravel(
6665
np.r_[compmodel(p[0], p[1:3], x1) - y1,
6766
compmodel(p[0], p[3:], x2) - y2])
@@ -72,7 +71,7 @@ def testJointFitter(self):
7271
@pytest.mark.skipif('not HAS_SCIPY')
7372
def test_no_constraints(self):
7473
g1 = models.Gaussian1DModel(9.9, 14.5, stddev=.3)
75-
func = lambda p, x: p[0]* np.exp((-(1/(p[2]**2)) * (x-p[1])**2))
74+
func = lambda p, x: p[0]* np.exp(-0.5 / p[2]**2 * (x-p[1])**2)
7675
errf = lambda p, x, y: func(p, x) - y
7776
p0 = [9.9, 14.5, 0.3]
7877
y = g1(self.x)

‎astropy/modeling/tests/test_fitters.py‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -109,7 +109,7 @@ def test_joint_fitter(self):
109109
p2 = [13, .4]
110110
A = 9.8
111111
p = np.r_[A, p1, p2]
112-
compmodel = lambda A, p, x: A* np.exp((-(1/(p[1]**2)) * (x-p[0])**2))
112+
compmodel = lambda A, p, x: A* np.exp(-0.5/p[1]**2 * (x-p[0])**2)
113113
errf = lambda p, x1, y1, x2, y2: np.ravel(np.r_[compmodel(p[0], p[1:3],
114114
x1) -y1, compmodel(p[0], p[3:], x2) - y2])
115115
coeff, _ = optimize.leastsq(errf, p, args=(self.x, self.ny1, self.x,
@@ -145,10 +145,10 @@ class TestNonLinearFitters(object):
145145
"""
146146
def setup_class(self):
147147
self.initial_values = [100, 5, 1]
148-
func = lambda p, x: p[0]* np.exp((-(1/(p[2]**2)) * (x-p[1])**2))
148+
func = lambda p, x: p[0]* np.exp(-0.5 / p[2]**2 * (x-p[1])**2)
149149
errf = lambda p, x, y: (func(p, x) - y)
150150
self.xdata = np.arange(0, 10, 0.1)
151-
sigma = 10 * np.ones_like(self.xdata)
151+
sigma = 8. * np.ones_like(self.xdata)
152152
rsn = RandomState(1234567890)
153153
yerror = rsn.normal(0, sigma)
154154
self.ydata = func(self.initial_values, self.xdata)+ yerror
@@ -167,7 +167,7 @@ def test_with_optimize(self):
167167
g1= models.Gaussian1DModel(100, 5, stddev=1, jacobian_func='estimated')
168168
fitter = fitting.NonLinearLSQFitter(g1)
169169
fitter(self.xdata, self.ydata)
170-
func = lambda p, x: p[0]* np.exp((-(1/(p[2]**2)) * (x-p[1])**2))
170+
func = lambda p, x: p[0]* np.exp(-0.5 / p[2]**2 * (x-p[1])**2)
171171
errf = lambda p, x, y: (func(p, x) - y)
172172
result = optimize.leastsq(errf, self.initial_values, args=(self.xdata, self.ydata))
173173
utils.assert_allclose(g1.parameters, result[0], rtol=10**(-3))

0 commit comments

Comments
 (0)