Mercurial > ift6266

comparison baseline/mlp/mlp_nist.py @ 414:3dba84c0fbc1

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
saving test score from best validation score in db now
author xaviermuller
date Thu, 29 Apr 2010 17:04:12 -0400
parents 195f95c3d461
children 868f82777839
comparison
equal deleted inserted replaced
413:f2dd75248483 414:3dba84c0fbc1
192 min_error_count=0.0 192 min_error_count=0.0
193 min_exemple_count=0.0 193 min_exemple_count=0.0
194 194
195 maj_error_count=0.0 195 maj_error_count=0.0
196 maj_exemple_count=0.0 196 maj_exemple_count=0.0
197
198 vtotal_error_count=0.0
199 vtotal_exemple_count=0.0
200
201 vnb_error_count=0.0
202 vnb_exemple_count=0.0
203
204 vchar_error_count=0.0
205 vchar_exemple_count=0.0
206
207 vmin_error_count=0.0
208 vmin_exemple_count=0.0
209
210 vmaj_error_count=0.0
211 vmaj_exemple_count=0.0
197 212
198 213
199 214
200 if data_set==0: 215 if data_set==0:
201 dataset=datasets.nist_all() 216 dataset=datasets.nist_all()
254 predicted_class=numpy.argmax(a1_out[36:62])+36 269 predicted_class=numpy.argmax(a1_out[36:62])+36
255 if(predicted_class!=wanted_class): 270 if(predicted_class!=wanted_class):
256 min_error_count = min_error_count +1 271 min_error_count = min_error_count +1
257 272
258 273
274
275 vtest_score=0
276 vtemp=0
277 for xt,yt in dataset.valid(1):
278
279 vtotal_exemple_count = vtotal_exemple_count +1
280 #get activation for layer 1
281 a0=numpy.dot(numpy.transpose(W1),numpy.transpose(xt[0])) + b1
282 #add non linear function to layer 1 activation
283 a0_out=numpy.tanh(a0)
284
285 #get activation for output layer
286 a1= numpy.dot(numpy.transpose(W2),a0_out) + b2
287 #add non linear function for output activation (softmax)
288 a1_exp = numpy.exp(a1)
289 sum_a1=numpy.sum(a1_exp)
290 a1_out=a1_exp/sum_a1
291
292 predicted_class=numpy.argmax(a1_out)
293 wanted_class=yt[0]
294 if(predicted_class!=wanted_class):
295 vtotal_error_count = vtotal_error_count +1
296
297 #treat digit error
298 if(wanted_class<10):
299 vnb_exemple_count=vnb_exemple_count + 1
300 predicted_class=numpy.argmax(a1_out[0:10])
301 if(predicted_class!=wanted_class):
302 vnb_error_count = vnb_error_count +1
303
304 if(wanted_class>9):
305 vchar_exemple_count=vchar_exemple_count + 1
306 predicted_class=numpy.argmax(a1_out[10:62])+10
307 if((predicted_class!=wanted_class) and ((predicted_class+26)!=wanted_class) and ((predicted_class-26)!=wanted_class)):
308 vchar_error_count = vchar_error_count +1
309
310 #minuscule
311 if(wanted_class>9 and wanted_class<36):
312 vmaj_exemple_count=vmaj_exemple_count + 1
313 predicted_class=numpy.argmax(a1_out[10:35])+10
314 if(predicted_class!=wanted_class):
315 vmaj_error_count = vmaj_error_count +1
316 #majuscule
317 if(wanted_class>35):
318 vmin_exemple_count=vmin_exemple_count + 1
319 predicted_class=numpy.argmax(a1_out[36:62])+36
320 if(predicted_class!=wanted_class):
321 vmin_error_count = vmin_error_count +1
322
259 323
260 print (('total error = %f') % ((total_error_count/total_exemple_count)*100.0)) 324 print (('total error = %f') % ((total_error_count/total_exemple_count)*100.0))
261 print (('number error = %f') % ((nb_error_count/nb_exemple_count)*100.0)) 325 print (('number error = %f') % ((nb_error_count/nb_exemple_count)*100.0))
262 print (('char error = %f') % ((char_error_count/char_exemple_count)*100.0)) 326 print (('char error = %f') % ((char_error_count/char_exemple_count)*100.0))
263 print (('min error = %f') % ((min_error_count/min_exemple_count)*100.0)) 327 print (('min error = %f') % ((min_error_count/min_exemple_count)*100.0))
264 print (('maj error = %f') % ((maj_error_count/maj_exemple_count)*100.0)) 328 print (('maj error = %f') % ((maj_error_count/maj_exemple_count)*100.0))
329
330 print (('valid total error = %f') % ((vtotal_error_count/vtotal_exemple_count)*100.0))
331 print (('valid number error = %f') % ((vnb_error_count/vnb_exemple_count)*100.0))
332 print (('valid char error = %f') % ((vchar_error_count/vchar_exemple_count)*100.0))
333 print (('valid min error = %f') % ((vmin_error_count/vmin_exemple_count)*100.0))
334 print (('valid maj error = %f') % ((vmaj_error_count/vmaj_exemple_count)*100.0))
335
336 print ((' num total = %d,%d') % (total_exemple_count,total_error_count))
337 print ((' num nb = %d,%d') % (nb_exemple_count,nb_error_count))
338 print ((' num min = %d,%d') % (min_exemple_count,min_error_count))
339 print ((' num maj = %d,%d') % (maj_exemple_count,maj_error_count))
340 print ((' num char = %d,%d') % (char_exemple_count,char_error_count))
265 return (total_error_count/total_exemple_count)*100.0 341 return (total_error_count/total_exemple_count)*100.0
266 342
267 343
268 344
269 345
290 configuration = [learning_rate,nb_max_exemples,nb_hidden,adaptive_lr] 366 configuration = [learning_rate,nb_max_exemples,nb_hidden,adaptive_lr]
291 367
292 #save initial learning rate if classical adaptive lr is used 368 #save initial learning rate if classical adaptive lr is used
293 initial_lr=learning_rate 369 initial_lr=learning_rate
294 max_div_count=1000 370 max_div_count=1000
371 optimal_test_error=0
295 372
296 373
297 total_validation_error_list = [] 374 total_validation_error_list = []
298 total_train_error_list = [] 375 total_train_error_list = []
299 learning_rate_list=[] 376 learning_rate_list=[]
480 print(('epoch %i, minibatch %i, test error of best ' 557 print(('epoch %i, minibatch %i, test error of best '
481 'model %f %%') % 558 'model %f %%') %
482 (epoch, minibatch_index+1, 559 (epoch, minibatch_index+1,
483 test_score*100.)) 560 test_score*100.))
484 sys.stdout.flush() 561 sys.stdout.flush()
562 optimal_test_error=test_score
485 563
486 # if the validation error is going up, we are overfitting (or oscillating) 564 # if the validation error is going up, we are overfitting (or oscillating)
487 # check if we are allowed to continue and if we will adjust the learning rate 565 # check if we are allowed to continue and if we will adjust the learning rate
488 elif this_validation_loss >= best_validation_loss: 566 elif this_validation_loss >= best_validation_loss:
489 567
549 #save the model and the weights 627 #save the model and the weights
550 numpy.savez('model.npy', config=configuration, W1=classifier.W1.value,W2=classifier.W2.value, b1=classifier.b1.value,b2=classifier.b2.value) 628 numpy.savez('model.npy', config=configuration, W1=classifier.W1.value,W2=classifier.W2.value, b1=classifier.b1.value,b2=classifier.b2.value)
551 numpy.savez('results.npy',config=configuration,total_train_error_list=total_train_error_list,total_validation_error_list=total_validation_error_list,\ 629 numpy.savez('results.npy',config=configuration,total_train_error_list=total_train_error_list,total_validation_error_list=total_validation_error_list,\
552 learning_rate_list=learning_rate_list, divergence_flag_list=divergence_flag_list) 630 learning_rate_list=learning_rate_list, divergence_flag_list=divergence_flag_list)
553 631
554 return (best_training_error*100.0,best_validation_loss * 100.,test_score*100.,best_iter*batch_size,(end_time-start_time)/60) 632 return (best_training_error*100.0,best_validation_loss * 100.,optimal_test_error*100.,best_iter*batch_size,(end_time-start_time)/60)
555 633
556 634
557 if __name__ == '__main__': 635 if __name__ == '__main__':
558 mlp_full_mnist() 636 mlp_full_mnist()
559 637