1 | #!/usr/bin/env python |
---|
2 | """Auxiliary numerical tools |
---|
3 | |
---|
4 | """ |
---|
5 | |
---|
6 | from math import acos, pi, sqrt |
---|
7 | from warnings import warn |
---|
8 | |
---|
9 | import anuga.utilities.log as log |
---|
10 | import numpy as num |
---|
11 | |
---|
12 | #After having migrated to numpy we should use the native NAN. |
---|
13 | #num.seterr(divide='warn') |
---|
14 | num.seterr(divide='ignore') # Ignore division error here for the time being |
---|
15 | NAN = (num.array([1])/0.)[0] |
---|
16 | |
---|
17 | # Static variable used by get_machine_precision |
---|
18 | machine_precision = None |
---|
19 | |
---|
20 | |
---|
21 | def safe_acos(x): |
---|
22 | """Safely compute acos |
---|
23 | |
---|
24 | Protect against cases where input argument x is outside the allowed |
---|
25 | interval [-1.0, 1.0] by no more than machine precision |
---|
26 | """ |
---|
27 | |
---|
28 | error_msg = 'Input to acos is outside allowed domain [-1.0, 1.0].'+\ |
---|
29 | 'I got %.12f' %x |
---|
30 | warning_msg = 'Changing argument to acos from %.18f to %.1f' %(x, sign(x)) |
---|
31 | |
---|
32 | eps = get_machine_precision() # Machine precision |
---|
33 | if x < -1.0: |
---|
34 | if x < -1.0 - eps: |
---|
35 | raise ValueError, errmsg |
---|
36 | else: |
---|
37 | warn(warning_msg) |
---|
38 | x = -1.0 |
---|
39 | |
---|
40 | if x > 1.0: |
---|
41 | if x > 1.0 + eps: |
---|
42 | raise ValueError, errmsg |
---|
43 | else: |
---|
44 | log.critical('NOTE: changing argument to acos from %.18f to 1.0' %x) |
---|
45 | x = 1.0 |
---|
46 | |
---|
47 | return acos(x) |
---|
48 | |
---|
49 | |
---|
50 | def sign(x): |
---|
51 | if x > 0: return 1 |
---|
52 | if x < 0: return -1 |
---|
53 | if x == 0: return 0 |
---|
54 | |
---|
55 | |
---|
56 | def is_scalar(x): |
---|
57 | """True if x is a scalar (constant numeric value) |
---|
58 | """ |
---|
59 | |
---|
60 | from types import IntType, FloatType |
---|
61 | if type(x) in [IntType, FloatType]: |
---|
62 | return True |
---|
63 | else: |
---|
64 | return False |
---|
65 | |
---|
66 | def angle(v1, v2=None): |
---|
67 | """Compute angle between 2D vectors v1 and v2. |
---|
68 | |
---|
69 | If v2 is not specified it will default |
---|
70 | to e1 (the unit vector in the x-direction) |
---|
71 | |
---|
72 | The angle is measured as a number in [0, 2pi] from v2 to v1. |
---|
73 | """ |
---|
74 | |
---|
75 | # Prepare two numeric vectors |
---|
76 | if v2 is None: |
---|
77 | v2 = [1.0, 0.0] # Unit vector along the x-axis |
---|
78 | |
---|
79 | v1 = ensure_numeric(v1, num.float) |
---|
80 | v2 = ensure_numeric(v2, num.float) |
---|
81 | |
---|
82 | # Normalise |
---|
83 | v1 = v1/num.sqrt(num.sum(v1**2)) |
---|
84 | v2 = v2/num.sqrt(num.sum(v2**2)) |
---|
85 | |
---|
86 | # Compute angle |
---|
87 | p = num.inner(v1, v2) |
---|
88 | c = num.inner(v1, normal_vector(v2)) # Projection onto normal |
---|
89 | # (negative cross product) |
---|
90 | |
---|
91 | theta = safe_acos(p) |
---|
92 | |
---|
93 | |
---|
94 | # Correct if v1 is in quadrant 3 or 4 with respect to v2 (as the x-axis) |
---|
95 | # If v2 was the unit vector [1,0] this would correspond to the test |
---|
96 | # if v1[1] < 0: theta = 2*pi-theta |
---|
97 | # In general we use the sign of the projection onto the normal. |
---|
98 | if c < 0: |
---|
99 | #Quadrant 3 or 4 |
---|
100 | theta = 2*pi-theta |
---|
101 | |
---|
102 | return theta |
---|
103 | |
---|
104 | |
---|
105 | def anglediff(v0, v1): |
---|
106 | """Compute difference between angle of vector v0 (x0, y0) and v1 (x1, y1). |
---|
107 | This is used for determining the ordering of vertices, |
---|
108 | e.g. for checking if they are counter clockwise. |
---|
109 | |
---|
110 | Always return a positive value |
---|
111 | """ |
---|
112 | |
---|
113 | from math import pi |
---|
114 | |
---|
115 | a0 = angle(v0) |
---|
116 | a1 = angle(v1) |
---|
117 | |
---|
118 | #Ensure that difference will be positive |
---|
119 | if a0 < a1: |
---|
120 | a0 += 2*pi |
---|
121 | |
---|
122 | return a0-a1 |
---|
123 | |
---|
124 | def normal_vector(v): |
---|
125 | """Normal vector to v. |
---|
126 | |
---|
127 | Returns vector 90 degrees counter clockwise to and of same length as v |
---|
128 | """ |
---|
129 | |
---|
130 | return num.array([-v[1], v[0]], num.float) |
---|
131 | |
---|
132 | |
---|
133 | #def crossproduct_length(v1, v2): |
---|
134 | # return v1[0]*v2[1]-v2[0]*v1[1] |
---|
135 | |
---|
136 | |
---|
137 | def mean(x): |
---|
138 | """Mean value of a vector |
---|
139 | """ |
---|
140 | return(float(num.sum(x))/len(x)) |
---|
141 | |
---|
142 | |
---|
143 | def cov(x, y=None): |
---|
144 | """Covariance of vectors x and y. |
---|
145 | |
---|
146 | If y is None: return cov(x, x) |
---|
147 | """ |
---|
148 | |
---|
149 | if y is None: |
---|
150 | y = x |
---|
151 | |
---|
152 | x = ensure_numeric(x) |
---|
153 | y = ensure_numeric(y) |
---|
154 | msg = 'Lengths must be equal: len(x) == %d, len(y) == %d' %(len(x), len(y)) |
---|
155 | assert(len(x)==len(y)), msg |
---|
156 | |
---|
157 | N = len(x) |
---|
158 | cx = x - mean(x) |
---|
159 | cy = y - mean(y) |
---|
160 | |
---|
161 | p = num.inner(cx,cy)/N |
---|
162 | return(p) |
---|
163 | |
---|
164 | |
---|
165 | def err(x, y=0, n=2, relative=True): |
---|
166 | """Relative error of ||x-y|| to ||y|| |
---|
167 | n = 2: Two norm |
---|
168 | n = None: Max norm |
---|
169 | |
---|
170 | If denominator evaluates to zero or |
---|
171 | if y is omitted or |
---|
172 | if keyword relative is False, |
---|
173 | absolute error is returned |
---|
174 | |
---|
175 | If there is x and y, n=2 and relative=False, this will calc; |
---|
176 | sqrt(sum_over_x&y((xi - yi)^2)) |
---|
177 | |
---|
178 | Given this value (err), to calc the root mean square deviation, do |
---|
179 | err/sqrt(n) |
---|
180 | where n is the number of elements,(len(x)) |
---|
181 | """ |
---|
182 | |
---|
183 | x = ensure_numeric(x) |
---|
184 | if y: |
---|
185 | y = ensure_numeric(y) |
---|
186 | |
---|
187 | if n == 2: |
---|
188 | err = norm(x-y) |
---|
189 | if relative is True: |
---|
190 | try: |
---|
191 | err = err/norm(y) |
---|
192 | except: |
---|
193 | pass |
---|
194 | |
---|
195 | else: |
---|
196 | err = max(abs(x-y)) |
---|
197 | if relative is True: |
---|
198 | try: |
---|
199 | err = err/max(abs(y)) |
---|
200 | except: |
---|
201 | pass |
---|
202 | |
---|
203 | return err |
---|
204 | |
---|
205 | |
---|
206 | def norm(x): |
---|
207 | """2-norm of x |
---|
208 | """ |
---|
209 | |
---|
210 | y = num.ravel(x) |
---|
211 | p = num.sqrt(num.inner(y,y)) |
---|
212 | return p |
---|
213 | |
---|
214 | |
---|
215 | def corr(x, y=None): |
---|
216 | """Correlation of x and y |
---|
217 | If y is None return autocorrelation of x |
---|
218 | """ |
---|
219 | |
---|
220 | from math import sqrt |
---|
221 | if y is None: |
---|
222 | y = x |
---|
223 | |
---|
224 | varx = cov(x) |
---|
225 | vary = cov(y) |
---|
226 | |
---|
227 | if varx == 0 or vary == 0: |
---|
228 | C = 0 |
---|
229 | else: |
---|
230 | C = cov(x,y)/sqrt(varx * vary) |
---|
231 | |
---|
232 | return(C) |
---|
233 | |
---|
234 | |
---|
235 | |
---|
236 | ## |
---|
237 | # @brief Ensure that a sequence is a numeric array of the required type. |
---|
238 | # @param A The sequence object to convert to a numeric array. |
---|
239 | # @param typecode The required numeric type of object A (a numeric dtype). |
---|
240 | # @return A numeric array of the required type. |
---|
241 | def ensure_numeric(A, typecode=None): |
---|
242 | """Ensure that sequence is a numeric array. |
---|
243 | |
---|
244 | Inputs: |
---|
245 | A: Sequence. If A is already a numeric array it will be returned |
---|
246 | unaltered |
---|
247 | If not, an attempt is made to convert it to a numeric |
---|
248 | array |
---|
249 | A: Scalar. Return 0-dimensional array containing that value. Note |
---|
250 | that a 0-dim array DOES NOT HAVE A LENGTH UNDER numpy. |
---|
251 | A: String. Array of ASCII values (numpy can't handle this) |
---|
252 | |
---|
253 | typecode: numeric type. If specified, use this in the conversion. |
---|
254 | If not, let numeric package decide. |
---|
255 | typecode will always be one of num.float, num.int, etc. |
---|
256 | |
---|
257 | Note that num.array(A, dtype) will sometimes copy. Use 'copy=False' to |
---|
258 | copy only when required. |
---|
259 | |
---|
260 | This function is necessary as array(A) can cause memory overflow. |
---|
261 | """ |
---|
262 | |
---|
263 | # if isinstance(A, basestring): |
---|
264 | # msg = 'Sorry, cannot handle strings in ensure_numeric()' |
---|
265 | # raise Exception, msg |
---|
266 | |
---|
267 | if typecode is None: |
---|
268 | if isinstance(A, num.ndarray): |
---|
269 | return A |
---|
270 | else: |
---|
271 | return num.array(A) |
---|
272 | else: |
---|
273 | return num.array(A, dtype=typecode, copy=False) |
---|
274 | |
---|
275 | |
---|
276 | def histogram(a, bins, relative=False): |
---|
277 | """Standard histogram straight from the numeric manual |
---|
278 | |
---|
279 | If relative is True, values will be normalised againts the total and |
---|
280 | thus represent frequencies rather than counts. |
---|
281 | """ |
---|
282 | |
---|
283 | n = num.searchsorted(num.sort(a), bins) |
---|
284 | n = num.concatenate([n, [len(a)]], axis=0) #??default# |
---|
285 | |
---|
286 | hist = n[1:]-n[:-1] |
---|
287 | |
---|
288 | if relative is True: |
---|
289 | hist = hist/float(num.sum(hist)) |
---|
290 | |
---|
291 | return hist |
---|
292 | |
---|
293 | def create_bins(data, number_of_bins = None): |
---|
294 | """Safely create bins for use with histogram |
---|
295 | If data contains only one point or is constant, one bin will be created. |
---|
296 | If number_of_bins in omitted 10 bins will be created |
---|
297 | """ |
---|
298 | |
---|
299 | mx = max(data) |
---|
300 | mn = min(data) |
---|
301 | |
---|
302 | if mx == mn: |
---|
303 | bins = num.array([mn]) |
---|
304 | else: |
---|
305 | if number_of_bins is None: |
---|
306 | number_of_bins = 10 |
---|
307 | |
---|
308 | bins = num.arange(mn, mx, (mx-mn)/number_of_bins) |
---|
309 | |
---|
310 | return bins |
---|
311 | |
---|
312 | |
---|
313 | |
---|
314 | def get_machine_precision(): |
---|
315 | """Calculate the machine precision for Floats |
---|
316 | |
---|
317 | Depends on static variable machine_precision in this module |
---|
318 | as this would otherwise require too much computation. |
---|
319 | """ |
---|
320 | |
---|
321 | global machine_precision |
---|
322 | |
---|
323 | if machine_precision is None: |
---|
324 | epsilon = 1. |
---|
325 | while epsilon/2 + 1. > 1.: |
---|
326 | epsilon /= 2 |
---|
327 | |
---|
328 | machine_precision = epsilon |
---|
329 | |
---|
330 | return machine_precision |
---|
331 | |
---|
332 | #################################################################### |
---|
333 | #Python versions of function that are also implemented in numerical_tools_ext.c |
---|
334 | # FIXME (Ole): Delete these and update tests |
---|
335 | # |
---|
336 | |
---|
337 | def gradient_python(x0, y0, x1, y1, x2, y2, q0, q1, q2): |
---|
338 | """ |
---|
339 | """ |
---|
340 | |
---|
341 | det = (y2-y0)*(x1-x0) - (y1-y0)*(x2-x0) |
---|
342 | a = (y2-y0)*(q1-q0) - (y1-y0)*(q2-q0) |
---|
343 | a /= det |
---|
344 | |
---|
345 | b = (x1-x0)*(q2-q0) - (x2-x0)*(q1-q0) |
---|
346 | b /= det |
---|
347 | |
---|
348 | return a, b |
---|
349 | |
---|
350 | |
---|
351 | def gradient2_python(x0, y0, x1, y1, q0, q1): |
---|
352 | """Compute radient based on two points and enforce zero gradient |
---|
353 | in the direction orthogonal to (x1-x0), (y1-y0) |
---|
354 | """ |
---|
355 | |
---|
356 | #Old code |
---|
357 | #det = x0*y1 - x1*y0 |
---|
358 | #if det != 0.0: |
---|
359 | # a = (y1*q0 - y0*q1)/det |
---|
360 | # b = (x0*q1 - x1*q0)/det |
---|
361 | |
---|
362 | #Correct code (ON) |
---|
363 | det = (x1-x0)**2 + (y1-y0)**2 |
---|
364 | if det != 0.0: |
---|
365 | a = (x1-x0)*(q1-q0)/det |
---|
366 | b = (y1-y0)*(q1-q0)/det |
---|
367 | |
---|
368 | return a, b |
---|
369 | |
---|
370 | ################################################################################ |
---|
371 | # Decision functions for numeric package objects. |
---|
372 | # It is a little tricky to decide if a numpy thing is of type float. |
---|
373 | # These functions hide numpy-specific details of how we do this. |
---|
374 | ################################################################################ |
---|
375 | |
---|
376 | ## |
---|
377 | # @brief Decide if an object is a numeric package object with datatype of float. |
---|
378 | # @param obj The object to decide on. |
---|
379 | # @return True if 'obj' is a numeric package object, and some sort of float. |
---|
380 | def is_num_float(obj): |
---|
381 | '''Is an object a numeric package float object?''' |
---|
382 | |
---|
383 | try: |
---|
384 | return obj.dtype.char in num.typecodes['Float'] |
---|
385 | except AttributeError: |
---|
386 | return False |
---|
387 | |
---|
388 | ## |
---|
389 | # @brief Decide if an object is a numeric package object with datatype of int. |
---|
390 | # @param obj The object to decide on. |
---|
391 | # @return True if 'obj' is a numeric package object, and some sort of int. |
---|
392 | def is_num_int(obj): |
---|
393 | '''Is an object a numeric package int object?''' |
---|
394 | |
---|
395 | try: |
---|
396 | return obj.dtype.char in num.typecodes['Integer'] |
---|
397 | except AttributeError: |
---|
398 | return False |
---|
399 | |
---|
400 | |
---|
401 | #----------------- |
---|
402 | #Initialise module |
---|
403 | |
---|
404 | from anuga.utilities import compile |
---|
405 | if compile.can_use_C_extension('util_ext.c'): |
---|
406 | from util_ext import gradient, gradient2 |
---|
407 | else: |
---|
408 | gradient = gradient_python |
---|
409 | gradient2 = gradient2_python |
---|
410 | |
---|
411 | |
---|
412 | if __name__ == '__main__': |
---|
413 | pass |
---|
414 | |
---|
415 | |
---|