if np.isnan(new_cost):
print index, "nan"
epoch_costs.append(new_cost)
#validation
if count == len(train_data):
valLoss = []
for i in range(len(dev_data)):
devInput = create_input(dev_data[i], parameters, True,
None, False)
newDevCost = f_test(*devInput)
valLoss.append(newDevCost)
#res_dev = evaluate(parameters, f_eval, dev_sentences,
# dev_data, id_to_tag, dico_tags,
# folder_out+fileNamePrefix+'.dev.txt',
# useAttend=False)
eStop.on_epoch_end(epoch, np.mean(valLoss))
if eStop.stop_training:
break
print BASH_YELLOW+"avg error:"+BASH_CLEAR,np.mean(epoch_costs),\
"avg dev loss:",np.mean(valLoss)
print BASH_YELLOW + "One epch espliced:" + BASH_CLEAR, time.time(
) - startTime
#start evaluate on test
res_test = evaluate(parameters,
f_eval,
test_sentences,
test_data,
id_to_tag,
dico_tags,
folder_out + fileNamePrefix + '.test.txt',
if eStop.stop_training:
break;
print "Starting epoch %i..." % epoch
for i, index in enumerate(np.random.permutation(len(train_data))):
count += 1
input = create_input(train_data[index], parameters, True, singletons)
new_cost = f_train(*input)
if np.isnan(new_cost):
print index,"nan"
epoch_costs.append(new_cost)
#validation
res_dev = evaluate(parameters, f_eval, dev_sentences,
dev_data, id_to_tag, dico_tags,
folder_out+fileNamePrefix+'.dev.txt')
eStop.on_epoch_end(epoch,res_dev['f1']) ;
print BASH_YELLOW+"avg error:"+BASH_CLEAR,np.mean(epoch_costs),\
" dev F1:",res_dev['f1'];
print BASH_YELLOW+"One epch espliced:"+BASH_CLEAR,time.time()-startTime;
#start evaluate on test
res_test = evaluate(parameters, f_eval, test_sentences,
test_data, id_to_tag, dico_tags,
folder_out+fileNamePrefix+'.test.txt')
if saveModel:
print "Saving model to disk..."
model.save()
print BASH_RED+'TEST: epoch'+BASH_CLEAR, epoch, 'F1', res_test['f1'],'p:',res_test['p'],'r:',res_test['r'], ' '*15
print BASH_YELLOW+"model important point:"+BASH_CLEAR,opts.String;
#}}}
else: