hosseinshn / basic-multi-task-learning Goto Github PK

if epoch % 1 == 0: print('Epoch: %d, Train Cost: %5.3f ' % (epoch, epoch_cost)) dv_cost = sess.run(cost, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5}) dv_cost1 = sess.run(cost1, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5}) cost1D.append(dv_cost1) dv_cost2 = sess.run(cost2, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5}) cost2D.append(dv_cost2)
Dropout should not be used in the evaluation, so should you change it to the following?
if epoch % 1 == 0: print('Epoch: %d, Train Cost: %5.3f ' % (epoch, epoch_cost)) dv_cost = sess.run(cost, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, **kp: 1**}) dv_cost1 = sess.run(cost1, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, **kp: 1**}) cost1D.append(dv_cost1) dv_cost2 = sess.run(cost2, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, **kp: 1**}) cost2D.append(dv_cost2)

Hi, thank you for sharing the multi-task learning code.
I have a question about the test step.
As described in the PyTorch code, there is a test dataset separated from the training and validation dataset.
I understood that training and validation sets used for model learning, but how can I test the model with a test dataset?
Do you have any suggestions or plans for update the code for this job?
I look forward to your reply.