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
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
|
module Lib
module Util
def self.compute_variance( old_variance, n, new_mean, old_mean, new_value )
# use revursiv formular for computing the variance
# ( see Tysiak, Folgen: explizit und rekursiv, ISSN: 0025-5866
# http://www.frl.de/tysiakpapers/07_TY_Papers.pdf )
return (n>1 ? old_variance * (n-2)/(n-1) : 0) +
(new_mean - old_mean)**2 +
(n>1 ? (new_value - new_mean)**2/(n-1) : 0 )
end
end
class Predictions
def identifier(instance_index)
return instance_index.to_s
end
def initialize( predicted_values,
actual_values,
confidence_values,
is_classification,
prediction_feature_values=nil )
@predicted_values = predicted_values
@actual_values = actual_values
@confidence_values = confidence_values
@is_classification = is_classification
@prediction_feature_values = prediction_feature_values
@num_classes = 1
#puts "predicted: "+predicted_values.inspect
#puts "actual: "+actual_values.inspect
#puts "confidence: "+confidence_values.inspect
raise "no predictions" if @predicted_values.size == 0
num_info = "predicted:"+@predicted_values.size.to_s+
" confidence:"+@confidence_values.size.to_s+" actual:"+@actual_values.size.to_s
raise "illegal num actual values "+num_info if @actual_values.size != @predicted_values.size
raise "illegal num confidence values "+num_info if @confidence_values.size != @predicted_values.size
@confidence_values.each{ |c| raise "illegal confidence value: '"+c.to_s+"'" unless c==nil or (c.is_a?(Numeric) and c>=0 and c<=1) }
conf_val_tmp = {}
@confidence_values.each{ |c| conf_val_tmp[c] = nil }
if conf_val_tmp.keys.size<2
LOGGER.warn("prediction w/o confidence values");
@confidence_values=nil
end
if @is_classification
raise "prediction_feature_values missing while performing classification" unless @prediction_feature_values
@num_classes = @prediction_feature_values.size
raise "num classes < 2" if @num_classes<2
{ "predicted"=>@predicted_values, "actual"=>@actual_values }.each do |s,values|
values.each{ |v| raise "illegal "+s+" classification-value ("+v.to_s+"),"+
"has to be either nil or index of predicted-values" if v!=nil and (v<0 or v>@num_classes)}
end
else
raise "prediction_feature_values != nil while performing regression" if @prediction_feature_values
{ "predicted"=>@predicted_values, "actual"=>@actual_values }.each do |s,values|
values.each{ |v| raise "illegal "+s+" regression-value ("+v.to_s+"),"+
"has to be either nil or number" unless v==nil or v.is_a?(Numeric)}
end
end
init_stats()
(0..@predicted_values.size-1).each do |i|
update_stats( @predicted_values[i], @actual_values[i], (@confidence_values!=nil)?@confidence_values[i]:nil )
end
end
def init_stats
@num_no_actual_value = 0
@num_with_actual_value = 0
@num_predicted = 0
@num_unpredicted = 0
if @is_classification
@confusion_matrix = []
@prediction_feature_values.each do |v|
@confusion_matrix.push( Array.new( @num_classes, 0 ) )
end
@num_correct = 0
@num_incorrect = 0
else
@sum_error = 0
@sum_abs_error = 0
@sum_squared_error = 0
@prediction_mean = 0
@actual_mean = 0
@variance_predicted = 0
@variance_actual = 0
@sum_actual = 0
@sum_predicted = 0
@sum_multiply = 0
@sum_squares_actual = 0
@sum_squares_predicted = 0
end
end
def update_stats( predicted_value, actual_value, confidence_value )
if actual_value==nil
@num_no_actual_value += 1
else
@num_with_actual_value += 1
if predicted_value==nil
@num_unpredicted += 1
else
@num_predicted += 1
if @is_classification
@confusion_matrix[actual_value][predicted_value] += 1
if (predicted_value == actual_value)
@num_correct += 1
else
@num_incorrect += 1
end
else
delta = predicted_value - actual_value
@sum_error += delta
@sum_abs_error += delta.abs
@sum_squared_error += delta**2
old_prediction_mean = @prediction_mean
@prediction_mean = (@prediction_mean * (@num_predicted-1) + predicted_value) / @num_predicted.to_f
old_actual_mean = @actual_mean
@actual_mean = (@actual_mean * (@num_predicted-1) + actual_value) / @num_predicted.to_f
@variance_predicted = Util.compute_variance( @variance_predicted, @num_predicted,
@prediction_mean, old_prediction_mean, predicted_value )
@variance_actual = Util.compute_variance( @variance_actual, @num_predicted,
@actual_mean, old_actual_mean, actual_value )
@sum_actual += actual_value
@sum_predicted += predicted_value
@sum_multiply += (actual_value*predicted_value)
@sum_squares_actual += actual_value**2
@sum_squares_predicted += predicted_value**2
end
end
end
end
def percent_correct
raise "no classification" unless @is_classification
return 0 if @num_with_actual_value==0
return 100 * @num_correct / @num_with_actual_value.to_f
end
def percent_incorrect
raise "no classification" unless @is_classification
return 0 if @num_with_actual_value==0
return 100 * @num_incorrect / @num_with_actual_value.to_f
end
def accuracy
return percent_correct / 100.0
end
def percent_unpredicted
return 0 if @num_with_actual_value==0
return 100 * @num_unpredicted / @num_with_actual_value.to_f
end
def num_unpredicted
@num_unpredicted
end
def percent_without_class
return 0 if @predicted_values==0
return 100 * @num_no_actual_value / @predicted_values.size.to_f
end
def num_without_class
@num_no_actual_value
end
def num_correct
raise "no classification" unless @is_classification
return @num_correct
end
def num_incorrect
raise "no classification" unless @is_classification
return @num_incorrect
end
def num_unclassified
raise "no classification" unless @is_classification
return @num_unpredicted
end
# internal structure of confusion matrix:
# hash with keys: hash{ :confusion_matrix_actual => <class_value>, :confusion_matrix_predicted => <class_value> }
# and values: <int-value>
def confusion_matrix
raise "no classification" unless @is_classification
res = {}
(0..@num_classes-1).each do |actual|
(0..@num_classes-1).each do |predicted|
res[{:confusion_matrix_actual => @prediction_feature_values[actual],
:confusion_matrix_predicted => @prediction_feature_values[predicted]}] = @confusion_matrix[actual][predicted]
end
end
return res
end
def area_under_roc(class_index=nil)
return prediction_feature_value_map( lambda{ |i| area_under_roc(i) } ) if class_index==nil
return 0.0 if @confidence_values==nil
LOGGER.warn("TODO: implement approx computiation of AUC,"+
"so far Wilcoxon-Man-Whitney is used (exponential)") if @predicted_values.size>1000
tp_conf = []
fp_conf = []
(0..@predicted_values.size-1).each do |i|
if @predicted_values[i]==class_index
if @actual_values[i]==class_index
tp_conf.push(@confidence_values[i])
else
fp_conf.push(@confidence_values[i])
end
end
end
return 0.0 if tp_conf.size == 0
return 1.0 if fp_conf.size == 0
sum = 0
tp_conf.each do |tp|
fp_conf.each do |fp|
sum += 1 if tp>fp
end
end
return sum / (tp_conf.size * fp_conf.size).to_f
end
def f_measure(class_index=nil)
return prediction_feature_value_map( lambda{ |i| f_measure(i) } ) if class_index==nil
prec = precision(class_index)
rec = recall(class_index)
return 0 if prec == 0 and rec == 0
return 2 * prec * rec / (prec + rec).to_f;
end
def precision(class_index=nil)
return prediction_feature_value_map( lambda{ |i| precision(i) } ) if class_index==nil
correct = 0
total = 0
(0..@num_classes-1).each do |i|
correct += @confusion_matrix[i][class_index] if i == class_index
total += @confusion_matrix[i][class_index]
end
return 0 if total==0
return correct/total.to_f
end
def recall(class_index=nil)
return true_positive_rate(class_index)
end
def true_negative_rate(class_index=nil)
return prediction_feature_value_map( lambda{ |i| true_negative_rate(i) } ) if class_index==nil
correct = 0
total = 0
(0..@num_classes-1).each do |i|
if i != class_index
(0..@num_classes-1).each do |j|
correct += @confusion_matrix[i][j] if j != class_index
total += @confusion_matrix[i][j]
end
end
end
return 0 if total==0
return correct/total.to_f
end
def num_true_negatives(class_index=nil)
return prediction_feature_value_map( lambda{ |i| num_true_negatives(i) } ) if class_index==nil
correct = 0
(0..@num_classes-1).each do |i|
if i != class_index
(0..@num_classes-1).each do |j|
correct += @confusion_matrix[i][j] if j != class_index
end
end
end
return correct
end
def true_positive_rate(class_index=nil)
return prediction_feature_value_map( lambda{ |i| true_positive_rate(i) } ) if class_index==nil
correct = 0
total = 0
(0..@num_classes-1).each do |i|
correct += @confusion_matrix[class_index][i] if i == class_index
total += @confusion_matrix[class_index][i]
end
return 0 if total==0
return correct/total.to_f
end
def num_true_positives(class_index=nil)
return prediction_feature_value_map( lambda{ |i| num_true_positives(i) } ) if class_index==nil
correct = 0
(0..@num_classes-1).each do |i|
correct += @confusion_matrix[class_index][i] if i == class_index
end
return correct
end
def false_negative_rate(class_index=nil)
return prediction_feature_value_map( lambda{ |i| false_negative_rate(i) } ) if class_index==nil
total = 0
incorrect = 0
(0..@num_classes-1).each do |i|
if i == class_index
(0..@num_classes-1).each do |j|
incorrect += @confusion_matrix[i][j] if j != class_index
total += @confusion_matrix[i][j]
end
end
end
return 0 if total == 0
return incorrect / total.to_f
end
def num_false_negatives(class_index=nil)
return prediction_feature_value_map( lambda{ |i| num_false_negatives(i) } ) if class_index==nil
incorrect = 0
(0..@num_classes-1).each do |i|
if i == class_index
(0..@num_classes-1).each do |j|
incorrect += @confusion_matrix[i][j] if j != class_index
end
end
end
return incorrect
end
def false_positive_rate(class_index=nil)
return prediction_feature_value_map( lambda{ |i| false_positive_rate(i) } ) if class_index==nil
total = 0
incorrect = 0
(0..@num_classes-1).each do |i|
if i != class_index
(0..@num_classes-1).each do |j|
incorrect += @confusion_matrix[i][j] if j == class_index
total += @confusion_matrix[i][j]
end
end
end
return 0 if total == 0
return incorrect / total.to_f
end
def num_false_positives(class_index=nil)
return prediction_feature_value_map( lambda{ |i| num_false_positives(i) } ) if class_index==nil
incorrect = 0
(0..@num_classes-1).each do |i|
if i != class_index
(0..@num_classes-1).each do |j|
incorrect += @confusion_matrix[i][j] if j == class_index
end
end
end
return incorrect
end
def weighted_area_under_roc
return weighted_measure( :area_under_roc )
end
def weighted_f_measure
return weighted_measure( :f_measure )
end
private
def weighted_measure( measure )
sum_instances = 0
num_instances_per_class = Array.new(@num_classes, 0)
(0..@num_classes-1).each do |i|
(0..@num_classes-1).each do |j|
num_instances_per_class[i] += @confusion_matrix[i][j]
end
sum_instances += num_instances_per_class[i]
end
raise "sum instances ("+sum_instances.to_s+") != num predicted ("+@num_predicted.to_s+")" unless @num_predicted == sum_instances
weighted = 0;
(0..@num_classes-1).each do |i|
weighted += self.send(measure,i) * num_instances_per_class[i]
end
return weighted / @num_predicted.to_f
end
# regression #######################################################################################
public
def root_mean_squared_error
return 0 if (@num_with_actual_value - @num_unpredicted)==0
Math.sqrt(@sum_squared_error / (@num_with_actual_value - @num_unpredicted).to_f)
end
def mean_absolute_error
return 0 if (@num_with_actual_value - @num_unpredicted)==0
Math.sqrt(@sum_abs_error / (@num_with_actual_value - @num_unpredicted).to_f)
end
def sum_squared_error
return @sum_squared_error
end
def r_square
return sample_correlation_coefficient ** 2
end
def sample_correlation_coefficient
# formula see http://en.wikipedia.org/wiki/Correlation_and_dependence#Pearson.27s_product-moment_coefficient
return ( @num_predicted * @sum_multiply - @sum_actual * @sum_predicted ) /
( Math.sqrt( @num_predicted * @sum_squares_actual - @sum_actual**2 ) *
Math.sqrt( @num_predicted * @sum_squares_predicted - @sum_predicted**2 ) )
end
def total_sum_of_squares
return @variance_actual * ( @num_predicted - 1 )
end
def target_variance_predicted
return @variance_predicted
end
def target_variance_actual
return @variance_actual
end
# data for roc-plots ###################################################################################
def get_roc_values(class_value)
raise "no confidence values" if @confidence_values==nil
class_index = @prediction_feature_values.index(class_value)
raise "class not found "+class_value.to_s if class_index==nil and class_value!=nil
c = []; p = []; a = []
(0..@predicted_values.size-1).each do |i|
# NOTE: not predicted instances are ignored here
if (@predicted_values[i]!=nil and (class_value==nil or @predicted_values[i]==class_index))
c << @confidence_values[i]
p << @predicted_values[i]
a << @actual_values[i]
end
end
return {:predicted_values => p, :actual_values => a, :confidence_values => c}
end
########################################################################################
def num_instances
return @predicted_values.size
end
def predicted_values
@predicted_values
end
def predicted_value(instance_index)
if @is_classification
@predicted_values[instance_index]==nil ? nil : @prediction_feature_values[@predicted_values[instance_index]]
else
@predicted_values[instance_index]
end
end
def actual_values
@actual_values
end
def actual_value(instance_index)
if @is_classification
@actual_values[instance_index]==nil ? nil : @prediction_feature_values[@actual_values[instance_index]]
else
@actual_values[instance_index]
end
end
def confidence_value(instance_index)
return @confidence_values[instance_index]
end
def classification_miss?(instance_index)
raise "no classification" unless @is_classification
return false if predicted_value(instance_index)==nil or actual_value(instance_index)==nil
return predicted_value(instance_index) != actual_value(instance_index)
end
def classification?
@is_classification
end
def confidence_values_available?
return @confidence_values!=nil
end
###################################################################################################################
#def compound(instance_index)
#return "instance_index.to_s"
#end
private
def prediction_feature_value_map(proc)
res = {}
(0..@num_classes-1).each do |i|
res[@prediction_feature_values[i]] = proc.call(i)
end
return res
end
end
end
|