-
Notifications
You must be signed in to change notification settings - Fork 2
/
cfeval.py
390 lines (343 loc) · 9.67 KB
/
cfeval.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
# various evaluation measures for collaboration filtering
import numpy
from numpy import *
from scipy.sparse import *
'''
X is prediction, Y is ground truth
Both X and Y should be scipy.sparse.csc_matrix
'''
##########################
# signed RMSE
##########################
#def srmse(X, Y):
# signs = zeros(X.data.size)
# errs = X.data-Y.data
# signs[errs>0] = 1.0 # overestimation
# signs[errs<0] = -1.0 # underestimation
# errs = errs**2
# return sqrt(sum(multiply(signs,errs))/X.size)
def rmse(X, Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
if X.size > 0:
return numpy.sqrt(sum((X.data - Y.data) ** 2) / X.size)
else:
return 0
#############################
## signed MAE
#############################
#def smae(X, Y):
# signs = zeros(X.data.size)
# errs = X.data-Y.data
# signs[errs>0] = 1.0 # overestimation
# signs[errs<0] = -1.0 # underestimation
# errs = abs(errs)
# return sum(multiply(signs,errs))/X.size
def mae(X, Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
if X.size > 0:
return sum(abs(X.data - Y.data)) / X.size
else:
return 0
def map(X, Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if len(unique(Yi)) == 1:
continue
I = argsort(-Xi)
[inds1] = where(Yi[I] >= 1)
nvalid += 1
pres = numpy.divide(arange(1, inds1.size + 1), 1.0 + inds1) # to avoid integer arithmetic
res += mean(pres)
if nvalid > 0:
res = res / nvalid
else:
print "map warning! nvalid==0"
return res
###############################################################################
# a version of map based on ratings data that treat ratings
# greater than r0 as relevant and all others as irreleant
###############################################################################
def map_rating(X,Y,r0):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
assert(r0>0)
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if all(Yi<r0):
continue
I = argsort(-Xi)
[inds1] = where(Yi[I] >= r0)
nvalid += 1
pres = numpy.divide(arange(1, inds1.size + 1), 1.0 + inds1) # to avoid integer arithmetic
res += mean(pres)
if nvalid > 0:
res = res / nvalid
else:
print "map warning! nvalid==0"
return res
#################################
# hit ratio of top-k list
#################################
def f1_topk(Xpred, Xtst, r0, k):
M,N = Xtst.shape
ys = zeros(M)
nvalid = 0
res = 0
for i in xrange(N):
ni = Xpred.indptr[i+1]-Xpred.indptr[i]
assert(ni>=k)
xs_i = Xpred.data[Xpred.indptr[i]:Xpred.indptr[i]+k]
ids_i = Xpred.indices[Xpred.indptr[i]:Xpred.indptr[i]+k]
assert(all(xs_i[1:k]-xs_i[0:k-1]<=0))
ys[:] = 0
ys[Xtst.indices[Xtst.indptr[i]:Xtst.indptr[i+1]]] = Xtst.data[Xtst.indptr[i]:Xtst.indptr[i+1]]
n1_total = sum(ys>=r0)
if n1_total==0:
continue
nvalid += 1
ys_i = ys[ids_i]
n1_topk = sum(ys_i>=r0)
pre = float(n1_topk)/float(k)
rec = float(n1_topk)/float(n1_total)
if pre + rec > 0:
res += 2*pre*rec/(pre+rec)
if nvalid > 0:
res /= nvalid
return res
#################################
# hit ratio of top-k list
#################################
def hit_topk(Xpred, Xtst, r0, k):
M,N = Xtst.shape
ys = zeros(M)
nvalid = 0
res = 0
for i in xrange(N):
ni = Xpred.indptr[i+1]-Xpred.indptr[i]
assert(ni>=k)
xs_i = Xpred.data[Xpred.indptr[i]:Xpred.indptr[i]+k]
ids_i = Xpred.indices[Xpred.indptr[i]:Xpred.indptr[i]+k]
assert(all(xs_i[1:k]-xs_i[0:k-1]<=0))
ys[:] = 0
ys[Xtst.indices[Xtst.indptr[i]:Xtst.indptr[i+1]]] = Xtst.data[Xtst.indptr[i]:Xtst.indptr[i+1]]
n1_total = sum(ys>=r0)
if n1_total==0:
continue
nvalid += 1
ys_i = ys[ids_i]
n1_topk = sum(ys_i>=r0)
res += float(n1_topk)/float(k)
if nvalid > 0:
res /= nvalid
return res
# do not compute average precision for each
def mpr(X, Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if len(unique(Yi)) == 1:
continue
I = argsort(-Xi)
[inds1] = where(Yi[I] >= 1)
nvalid += inds1.size
pres = numpy.divide(arange(1, inds1.size + 1), 1.0 + inds1) # to avoid integer arithmetic
res += sum(pres)
assert(nvalid > 0)
res = res / nvalid
return res
# mean rank predicion, for each relevant item, we compute the proportion of
# irrelevant items ranked below it
def mrpr(X, Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if len(unique(Yi)) == 1: # must have multiple rating classes
continue
I = argsort(-Xi)
[inds1] = where(Yi[I] >= 1)
n0 = j1-j0-inds1.size # total number of irrelevant items
nvalid += inds1.size
rpr = 1-numpy.divide(inds1-arange(inds1.size), float(n0)) # to avoid integer arithmetic
res += sum(rpr)
assert(nvalid > 0)
res = res / nvalid
return res
####################################
# mean reciprocal rank for implicit
####################################
def mrr(X,Y):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if len(unique(Yi)) == 1: # must have multiple rating classes
continue
I = argsort(-Xi)
[inds1] = where(Yi[I] >= 1)
nvalid += inds1.size
res += sum(1.0/(1+inds1))
assert(nvalid > 0)
res = res / nvalid
return res
def ap_global(X,Y):
Xdata = X.data
Ydata = Y.data
if all(Ydata < 1):
assert False, "no relevant items found in the test data"
I = argsort(-Xdata)
[inds1] = where(Ydata[I] >= 1)
pres = numpy.divide(arange(1, inds1.size + 1), 1.0 + inds1) # to avoid integer arithmetic
return mean(pres)
def ndcgk_global(X,Y,K):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
(M,N) = Y.shape
Xdata = X.data
Ydata = Y.data
nnz = Ydata.size
I = argsort(-Xdata)
Y_pred = numpy.exp(Ydata[I])-1.0
Y_best = numpy.exp(-(sort(-Ydata)))-1.0
Wi = numpy.log(numpy.exp(1) + Ydata.size-1)
Yi_pred = numpy.divide(Y_pred, Wi)
Yi_best = numpy.divide(Y_best, Wi)
K = min([K, nnz])
res = sum(Yi_pred[0:K]) / sum(Yi_best[0:K])
return res
'''
Normalized Discounted Cummulative Gain at K
'''
def ndcg_k(X, Y, K):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = 0
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i+1]]
if j0 == j1: # skip empty column
continue
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
if all(Yi==0):
continue
nvalid += 1
I = argsort(-Xi)
Yi_pred = numpy.exp(Yi[I])-1.0
Yi_best = numpy.exp(-(sort(-Yi)))-1.0
Wi = numpy.log(numpy.exp(1) + arange(j1 - j0))
Yi_pred = numpy.divide(Yi_pred, Wi)
Yi_best = numpy.divide(Yi_best, Wi)
Ki = min([K, j1 - j0])
res += sum(Yi_pred[0:Ki]) / sum(Yi_best[0:Ki])
assert(nvalid > 0)
res /= nvalid
return res
def ndcg_multi(X, Y, Ks):
assert(X.size == Y.size and all(X.indices == Y.indices) and all(X.indptr == Y.indptr))
n = Y.shape[1]
res = zeros(len(Ks))
nvalid = 0
Xdata = X.data
Ydata = Y.data
indices = Y.indices
indptr = Y.indptr
for i in xrange(n):
[j0, j1] = [indptr[i], indptr[i + 1]]
if j0 == j1: # skip empty column
continue
nvalid += 1
Xi = Xdata[j0:j1]
Yi = Ydata[j0:j1]
I = argsort(-Xi)
Yi_pred = numpy.exp(Yi[I])-1.0
Yi_best = numpy.exp(-(sort(-Yi)))-1.0
Wi = numpy.log(numpy.exp(1) + arange(j1 - j0))
Yi_pred = numpy.divide(Yi_pred, Wi)
Yi_best = numpy.divide(Yi_best, Wi)
for k in xrange(len(Ks)):
K = Ks[k]
Ki = min([K, j1 - j0])
res[k] += sum(Yi_pred[0:Ki]) / sum(Yi_best[0:Ki])
assert(nvalid > 0)
res /= nvalid
return res
#####################################
# Test MAP
#####################################
def test_map():
import scipy.io
import time
import scipy
import scipy.sparse
matdat = scipy.io.loadmat('D:\\Users\\nliu\\PyCF\\map_test.mat')
Y = matdat['Ytest']
X = matdat['Xtest']
#X.data = scipy.random.rand(X.data.size)
# tic = time.time()
# res1 = map_old(X, Y)
# toc = time.time()
# print "res1 : %f , time : %f" % (res1, toc - tic)
tic = time.time()
res2 = map(X, Y)
toc = time.time()
print "res2 : %f , time : %f" % (res2, toc - tic)
if __name__ == "__main__":
#test_metrics()
pass