1 # Copyright (C) 2009-2010 W. Trevor King <wking@drexel.edu>
3 # This program is free software: you can redistribute it and/or modify
4 # it under the terms of the GNU General Public License as published by
5 # the Free Software Foundation, either version 3 of the License, or
6 # (at your option) any later version.
8 # This program is distributed in the hope that it will be useful,
9 # but WITHOUT ANY WARRANTY; without even the implied warranty of
10 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 # GNU General Public License for more details.
13 # You should have received a copy of the GNU General Public License
14 # along with this program. If not, see <http://www.gnu.org/licenses/>.
16 # The author may be contacted at <wking@drexel.edu> on the Internet, or
17 # write to Trevor King, Drexel University, Physics Dept., 3141 Chestnut St.,
18 # Philadelphia PA 19104, USA.
20 """Histogram generation and comparison.
28 _multiprocess_can_split_ = True
29 """Allow nosetests to split tests between processes.
33 class Histogram (object):
34 """A histogram with a flexible comparison method, `residual()`.
41 def calculate_bin_edges(self, data, bin_width):
44 >>> h.calculate_bin_edges(numpy.array([-7.5, 18.2, 4]), 10)
45 array([-10., 0., 10., 20.])
46 >>> h.calculate_bin_edges(numpy.array([-7.5, 18.2, 4, 20]), 10)
47 array([-10., 0., 10., 20.])
48 >>> h.calculate_bin_edges(numpy.array([0, 18.2, 4, 20]), 10)
49 array([ 0., 10., 20.])
50 >>> h.calculate_bin_edges(numpy.array([18.2, 4, 20]), 10)
51 array([ 0., 10., 20.])
52 >>> h.calculate_bin_edges(numpy.array([18.2, 20]), 10)
57 bin_start = m - m % bin_width
58 return numpy.arange(bin_start, M+bin_width, bin_width, dtype=data.dtype)
60 def from_data(self, data, bin_edges):
61 """Initialize from `data`.
63 All bins should be of equal width (so we can calculate which
64 bin a data point belongs to).
66 data = numpy.array(data)
67 self.bin_edges = numpy.array(bin_edges)
68 bin_width = self.bin_edges[1] - self.bin_edges[0]
70 bin_is = numpy.floor((data - self.bin_edges[0])/bin_width)
71 self.counts = numpy.zeros((len(self.bin_edges)-1,), dtype=numpy.int)
72 for i in range(len(self.counts)):
73 self.counts[i] = (bin_is == i).sum()
74 self.counts = numpy.array(self.counts)
75 self.total = float(len(data)) # some data might be outside the bins
76 self.mean = data.mean()
77 self.std_dev = data.std()
80 def from_stream(self, stream):
81 """Initialize from `stream`.
85 # comment and blank lines ignored
86 <bin_edge><whitespace><count>
89 `<bin_edge>` should mark the left-hand side of the bin, and
90 all bins should be of equal width (so we know where the last
95 >>> h.from_stream(StringIO.StringIO('''# Force (N)\\tUnfolding events
101 ['Force (N)', 'Unfolding events']
106 >>> h.bin_edges # doctest: +ELLIPSIS
107 [1.5e-10, 2.000...e-10, 2.500...e-10, 3e-10]
108 >>> h.probabilities # doctest: +ELLIPSIS
109 [0.181..., 0.727..., 0.0909...]
114 for line in stream.readlines():
116 if len(line) == 0 or line.startswith('#'):
117 if self.headings == None and line.startswith('#'):
118 line = line[len('#'):]
119 self.headings = [x.strip() for x in line.split('\t')]
120 continue # ignore blank lines and comments
122 bin_edge,count = line.split()
124 log().error('Unable to parse histogram line: "%s"' % line)
126 self.bin_edges.append(float(bin_edge))
127 self.counts.append(float(count))
128 bin_width = self.bin_edges[1] - self.bin_edges[0]
129 self.bin_edges.append(self.bin_edges[-1]+bin_width)
132 def to_stream(self, stream):
133 """Write to `stream` with the same format as `from_stream()`.
137 >>> h.headings = ['Force (N)', 'Unfolding events']
138 >>> h.bin_edges = [1.5e-10, 2e-10, 2.5e-10, 3e-10]
139 >>> h.counts = [10, 40, 5]
140 >>> h.to_stream(sys.stdout)
141 ... # doctest: +NORMALIZE_WHITESPACE, +REPORT_UDIFF
142 #Force (N)\tUnfolding events
147 if self.headings != None:
148 stream.write('#%s\n' % '\t'.join(self.headings))
149 for bin,count in zip(self.bin_edges, self.counts):
150 stream.write('%g\t%g\n' % (bin, count))
153 """Analyze `.counts` and `.bin_edges` if the raw data is missing.
155 Generate `.total`, `.mean`, and `.std_dev`, and run
158 bin_width = self.bin_edges[1] - self.bin_edges[0]
159 self.total = float(sum(self.counts))
161 for bin,count in zip(self.bin_edges, self.counts):
162 bin += bin_width / 2.0
163 self.mean += bin * count
164 self.mean /= float(self.total)
166 for bin,count in zip(self.bin_edges, self.counts):
167 bin += bin_width / 2.0
168 variance += (bin - self.mean)**2 * count
169 self.std_dev = numpy.sqrt(variance)
173 """Generate `.probabilities` from `.total` and `.counts`.
175 self.total = float(self.total)
176 self.probabilities = [count/self.total for count in self.counts]
178 def mean_residual(self, other):
179 return abs(other.mean - self.mean)
181 def std_dev_residual(self, other):
182 return abs(other.std_dev - self.std_dev)
184 def chi_squared_residual(self, other):
185 assert (self.bin_edges == other.bin_edges).all()
187 for probA,probB in zip(self.probabilities, other.probabilities):
188 residual += (probA-probB)**2 / probB
191 def jensen_shannon_residual(self, other):
192 assert (self.bin_edges == other.bin_edges).all()
195 Kullback-Leibler divergence for a single bin, with the
196 exception that we return 0 if q==0, rather than
197 exploding like d_KL should. We can do this because
198 the way d_KL_term is used in the Jensen-Shannon
199 divergence, if q (really m) == 0, then p also == 0,
200 and the Jensen-Shannon divergence for the entire term
203 if p==0 or q==0 or p==q:
205 return p * numpy.log2(p/q)
207 for probA,probB in zip(self.probabilities, other.probabilities):
208 m = (probA+probB) / 2.0
209 residual += 0.5*(d_KL_term(probA,m) + d_KL_term(probB,m))
212 def _method_to_type(self, name):
213 return name[:-len('_residual')].replace('_', '-')
215 def _type_to_method(self, name):
216 return '%s_residual' % name.replace('-', '_')
219 """Return a list of supported residual types.
221 return sorted([self._method_to_type(x)
222 for x in dir(self) if x.endswith('_residual')])
224 def residual(self, other, type='jensen-shannon'):
225 """Compare this histogram with `other`.
227 Supported comparison `type`\s may be found with `types()`:
231 ['chi-squared', 'jensen-shannon', 'mean', 'std-dev']
233 Selecting an invalid `type` raises an exception.
235 >>> h.residual(other=None, type='invalid-type')
236 Traceback (most recent call last):
238 AttributeError: 'Histogram' object has no attribute 'invalid_type_residual'
240 For residual types where there is a convention, this histogram
241 is treated as the experimental histogram and the other
242 histogram is treated as the theoretical one.
244 r_method = getattr(self, self._type_to_method(type))
245 return r_method(other)