Coverage for python/lsst/sims/maf/metrics/moSummaryMetrics.py : 10%

Hot-keys on this page
r m x p toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1import numpy as np
2import warnings
4from .moMetrics import BaseMoMetric
6__all__ = ['integrateOverH', 'ValueAtHMetric', 'MeanValueAtHMetric',
7 'MoCompletenessMetric', 'MoCompletenessAtTimeMetric']
10def integrateOverH(Mvalues, Hvalues, Hindex = 0.33):
11 """Function to calculate a metric value integrated over an Hrange, assuming a power-law distribution.
13 Parameters
14 ----------
15 Mvalues : numpy.ndarray
16 The metric values at each H value.
17 Hvalues : numpy.ndarray
18 The H values corresponding to each Mvalue (must be the same length).
19 Hindex : float, opt
20 The power-law index expected for the H value distribution.
21 Default is 0.33 (dN/dH = 10^(Hindex * H) ).
23 Returns
24 --------
25 numpy.ndarray
26 The integrated or cumulative metric values.
27 """
28 # Set expected H distribution.
29 # dndh = differential size distribution (number in this bin)
30 dndh = np.power(10., Hindex*(Hvalues-Hvalues.min()))
31 # dn = cumulative size distribution (number in this bin and brighter)
32 intVals = np.cumsum(Mvalues*dndh)/np.cumsum(dndh)
33 return intVals
36class ValueAtHMetric(BaseMoMetric):
37 """Return the metric value at a given H value.
39 Requires the metric values to be one-dimensional (typically, completeness values).
41 Parameters
42 ----------
43 Hmark : float, opt
44 The H value at which to look up the metric value. Default = 22.
45 """
46 def __init__(self, Hmark=22, **kwargs):
47 metricName = 'Value At H=%.1f' %(Hmark)
48 super(ValueAtHMetric, self).__init__(metricName=metricName, **kwargs)
49 self.Hmark = Hmark
51 def run(self, metricVals, Hvals):
52 # Check if desired H value is within range of H values.
53 if (self.Hmark < Hvals.min()) or (self.Hmark > Hvals.max()):
54 warnings.warn('Desired H value of metric outside range of provided H values.')
55 return None
56 if metricVals.shape[0] != 1:
57 warnings.warn('This is not an appropriate summary statistic for this data - need 1d values.')
58 return None
59 value = np.interp(self.Hmark, Hvals, metricVals[0])
60 return value
63class MeanValueAtHMetric(BaseMoMetric):
64 """Return the mean value of a metric at a given H.
66 Allows the metric values to be multi-dimensional (i.e. use a cloned H distribution).
68 Parameters
69 ----------
70 Hmark : float, opt
71 The H value at which to look up the metric value. Default = 22.
72 """
73 def __init__(self, Hmark=22, reduceFunc=np.mean, metricName=None, **kwargs):
74 if metricName is None:
75 metricName = 'Mean Value At H=%.1f' %(Hmark)
76 super(MeanValueAtHMetric, self).__init__(metricName=metricName, **kwargs)
77 self.Hmark = Hmark
78 self.reduceFunc = reduceFunc
80 def run(self, metricVals, Hvals):
81 # Check if desired H value is within range of H values.
82 if (self.Hmark < Hvals.min()) or (self.Hmark > Hvals.max()):
83 warnings.warn('Desired H value of metric outside range of provided H values.')
84 return None
85 value = np.interp(self.Hmark, Hvals, self.reduceFunc(metricVals.swapaxes(0, 1), axis=1))
86 return value
89class MoCompletenessMetric(BaseMoMetric):
90 """Calculate the fraction of the population that meets `threshold` value or higher.
91 This is equivalent to calculating the completeness (relative to the entire population) given
92 the output of a Discovery_N_Chances metric, or the fraction of the population that meets a given cutoff
93 value for Color determination metrics.
95 Any moving object metric that outputs a float value can thus have the 'fraction of the population'
96 with greater than X value calculated here, as a summary statistic.
98 Parameters
99 ----------
100 threshold : int, opt
101 Count the fraction of the population that exceeds this value. Default = 1.
102 nbins : int, opt
103 If the H values for the metric are not a cloned distribution, then split up H into this many bins.
104 Default 20.
105 minHrange : float, opt
106 If the H values for the metric are not a cloned distribution, then split up H into at least this
107 range (otherwise just use the min/max of the H values). Default 1.0
108 cumulative : bool, opt
109 If False, simply report the differential fractional value (or differential completeness).
110 If True, integrate over the H distribution (using IntegrateOverH) to report a cumulative fraction.
111 Default False.
112 Hindex : float, opt
113 Use Hindex as the power law to integrate over H, if cumulative is True. Default 0.3.
114 """
115 def __init__(self, threshold=1, nbins=20, minHrange=1.0, cumulative=False, Hindex=0.33, **kwargs):
116 if 'metricName' in kwargs:
117 metricName = kwargs.pop('metricName')
118 if metricName.startswith('Cumulative'):
119 self.cumulative=True
120 units = '<= H'
121 else:
122 self.cumulative=False
123 units = '@ H'
124 else:
125 self.cumulative = cumulative
126 if self.cumulative:
127 metricName = 'CumulativeCompleteness'
128 units = '<= H'
129 else:
130 metricName = 'DifferentialCompleteness'
131 units = '@ H'
132 super(MoCompletenessMetric, self).__init__(metricName=metricName, units=units, **kwargs)
133 self.threshold = threshold
134 # If H is not a cloned distribution, then we need to specify how to bin these values.
135 self.nbins = nbins
136 self.minHrange = minHrange
137 self.Hindex = Hindex
139 def run(self, metricValues, Hvals):
140 nSsos = metricValues.shape[0]
141 nHval = len(Hvals)
142 metricValH = metricValues.swapaxes(0, 1)
143 if nHval == metricValues.shape[1]:
144 # Hvals array is probably the same as the cloned H array.
145 completeness = np.zeros(len(Hvals), float)
146 for i, H in enumerate(Hvals):
147 completeness[i] = np.where(metricValH[i].filled(0) >= self.threshold)[0].size
148 completeness = completeness / float(nSsos)
149 else:
150 # The Hvals are spread more randomly among the objects (we probably used one per object).
151 hrange = Hvals.max() - Hvals.min()
152 minH = Hvals.min()
153 if hrange < self.minHrange:
154 hrange = self.minHrange
155 minH = Hvals.min() - hrange/2.0
156 stepsize = hrange / float(self.nbins)
157 bins = np.arange(minH, minH + hrange + stepsize/2.0, stepsize)
158 Hvals = bins[:-1]
159 n_all, b = np.histogram(metricValH[0], bins)
160 condition = np.where(metricValH[0] >= self.requiredChances)[0]
161 n_found, b = np.histogram(metricValH[0][condition], bins)
162 completeness = n_found.astype(float) / n_all.astype(float)
163 completeness = np.where(n_all==0, 0, completeness)
164 if self.cumulative:
165 completenessInt = integrateOverH(completeness, Hvals, self.Hindex)
166 summaryVal = np.empty(len(completenessInt), dtype=[('name', np.str_, 20), ('value', float)])
167 summaryVal['value'] = completenessInt
168 for i, Hval in enumerate(Hvals):
169 summaryVal['name'][i] = 'H <= %f' % (Hval)
170 else:
171 summaryVal = np.empty(len(completeness), dtype=[('name', np.str_, 20), ('value', float)])
172 summaryVal['value'] = completeness
173 for i, Hval in enumerate(Hvals):
174 summaryVal['name'][i] = 'H = %f' % (Hval)
175 return summaryVal
177class MoCompletenessAtTimeMetric(BaseMoMetric):
178 """Calculate the completeness (relative to the entire population) <= a given H as a function of time,
179 given the times of each discovery.
181 Input values of the discovery times can come from the Discovery_Time (child) metric or the
182 KnownObjects metric.
184 Parameters
185 ----------
186 times : numpy.ndarray like
187 The bins to distribute the discovery times into. Same units as the discovery time (typically MJD).
188 Hval : float, opt
189 The value of H to count completeness at (or cumulative completeness to).
190 Default None, in which case a value halfway through Hvals (the slicer H range) will be chosen.
191 cumulative : bool, opt
192 If True, calculate the cumulative completeness (completeness <= H).
193 If False, calculate the differential completeness (completeness @ H).
194 Default True.
195 Hindex : float, opt
196 Use Hindex as the power law to integrate over H, if cumulative is True. Default 0.3.
197 """
199 def __init__(self, times, Hval=None, cumulative=True, Hindex=0.33, **kwargs):
200 self.Hval = Hval
201 self.times = times
202 self.Hindex = Hindex
203 if 'metricName' in kwargs:
204 metricName = kwargs.pop('metricName')
205 if metricName.startswith('Differential'):
206 self.cumulative = False
207 self.metricName = metricName
208 else:
209 self.cumulative = True
210 self.metricName = metricName
211 else:
212 self.cumulative = cumulative
213 if self.cumulative:
214 self.metricName = 'CumulativeCompleteness@Time@H=%.2f' % self.Hval
215 else:
216 self.metricName = 'DifferentialCompleteness@Time@H=%.2f' % self.Hval
217 self._setLabels()
218 super(MoCompletenessAtTimeMetric, self).__init__(metricName=self.metricName, units=self.units,
219 **kwargs)
221 def _setLabels(self):
222 if self.Hval is not None:
223 if self.cumulative:
224 self.units = 'H <=%.1f' % (self.Hval)
225 else:
226 self.units = 'H = %.1f' % (self.Hval)
227 else:
228 self.units = 'H'
230 def run(self, discoveryTimes, Hvals):
231 if len(Hvals) != discoveryTimes.shape[1]:
232 warnings.warn("This summary metric expects cloned H distribution. Cannot calculate summary.")
233 return
234 nSsos = discoveryTimes.shape[0]
235 timesinH = discoveryTimes.swapaxes(0, 1)
236 completenessH = np.empty([len(Hvals), len(self.times)], float)
237 for i, H in enumerate(Hvals):
238 n, b = np.histogram(timesinH[i].compressed(), bins=self.times)
239 completenessH[i][0] = 0
240 completenessH[i][1:] = n.cumsum()
241 completenessH = completenessH / float(nSsos)
242 completeness = completenessH.swapaxes(0, 1)
243 if self.cumulative:
244 for i, t in enumerate(self.times):
245 completeness[i] = integrateOverH(completeness[i], Hvals)
246 # To save the summary statistic, we must pick out a given H value.
247 if self.Hval is None:
248 Hidx = len(Hvals) // 2
249 self.Hval = Hvals[Hidx]
250 self._setLabels()
251 else:
252 Hidx = np.where(np.abs(Hvals - self.Hval) == np.abs(Hvals - self.Hval).min())[0][0]
253 self.Hval = Hvals[Hidx]
254 self._setLabels()
255 summaryVal = np.empty(len(self.times), dtype=[('name', np.str_, 20), ('value', float)])
256 summaryVal['value'] = completeness[:, Hidx]
257 for i, time in enumerate(self.times):
258 summaryVal['name'][i] = '%s @ %.2f' % (self.units, time)
259 return summaryVal