start = pre_index
check_index = start + batch_size
if check_index > data_size:
# Start next epoch
start = 0
end = start + batch_size
return start, end
### End define function ###
### Read file of train data and label
oriTrData, o.inDim = myio.read_data_file(trDataFile,o.inDim)
oriTrLabel_tmp = myio.read_label_file(trLabelFile)
oriTrLabel = myio.dense_to_one_hot(oriTrLabel_tmp,o.numClass)
oriTrData, oriTrLabel=trainShuff(oriTrData, oriTrLabel) # shuffling
valInx = oriTrData.shape[0]/100*valRate
valData = oriTrData[0:valInx]
valLabel = oriTrLabel[0:valInx]
trData = oriTrData[valInx+1:oriTrData.shape[0]]
trLabel = oriTrLabel[valInx+1:oriTrLabel.shape[0]]
totalBatch = trData.shape[0]/miniBatch
### Main script ###
x = graph.get_tensor_by_name("x:0")
lab_y = graph.get_tensor_by_name("lab_y:0")
#mm_last = graph.get_tensor_by_name("mm_last:0")
#b_last = graph.get_tensor_by_name("b_last:0")
out_y = graph.get_tensor_by_name("out_y:0")
keepProb = graph.get_tensor_by_name("keepProb:0")
ce = graph.get_tensor_by_name("ce:0")
acc = graph.get_tensor_by_name("acc:0")
test_data, featdim = myio.read_data_file(data_file)
print 'LOG : predict probability using DNN-model'
with tf.device('/cpu:0'):
pred_data = sess.run(out_y, feed_dict={x: test_data, keepProb: 1.0})
if in_lab:
test_lab = myio.read_label_file(label_file)
test_lab_ot = myio.dense_to_one_hot(test_lab, 2)
pred_acc = sess.run(acc, feed_dict={out_y: pred_data, lab_y: test_lab_ot})
pred_ce = sess.run(ce, feed_dict={out_y: pred_data, lab_y: test_lab_ot})
print 'Results : '
print ' # of data = %d' % (pred_data.shape[0])
print ' average of cross entropy = %f' % (pred_ce)
print ' accuracy = %2.1f%%' % (pred_acc * 100)
print '### done\n'
if o.predprob != '':
print 'LOG : write predicted probability -> %s' % (o.predprob)
myio.write_predicted_prob(pred_data, o.predprob)
if o.predlab != '':