I have the following recommendations for improving DREAM_DM_starter_tf.py code of Tensorflow:   1. This code does not have any method for reading png files but only dicom formatted files. It would be nice to include an option to read png or jpg formats. 2. Some subjects in metadata have '.' (i.e. a dot symbol) as the cancer label. I am presuming that it means unknown. For the time being, I modified the code and took those as having label 0. The code can be updated to handle such cases and we should be informed on what to do with unknown labels. 3. The code was not able to finish model training in the Express lane when I used GoogleNet, LeNet and AlexNet with the default parameters given in the docker example. I am receiving "Submission exceeded alloted time" message. I am not sure if the code can be made faster or the time limit of the Model training Express lane could be increased. Here are the command-lines I used for these networks in my train.sh:   GoogleNet python DREAM_DM_starter_tf.py \--lr 0.0001 \--reg 0.0001 \--decay 0.985 \--bs 50 \--net GoogLe \--ms 224 \--dropout 0.6   LeNet: python DREAM_DM_starter_tf.py \--lr 0.0001 \--reg 0.0001 \--decay 0.985 \--bs 10 \--net Le \--ms 32 \--epoch 2 \--dropout 0.5 \--saver "/modelState/model_le_1.ckpt"   AlexNet: python DREAM_DM_starter_tf.py \--lr 0.0001 \--reg 0.0001 \--decay 0.985 \--bs 10 \--net Alex \--ms 224 \--epoch 3 \--dropout 0.5 \--saver "/modelState/model_alex_1.ckpt" 4. I received the following error message for VGG16 in the Express Lane for model training:   STDOUT: Let's start the training! STDOUT: Network: VGG16, Dropout: 0.5, Reg: 0.0005, LR: 0.0001, Decay: 0.985 STDERR: Traceback (most recent call last): STDERR: File "DREAM_DM_starter_tf.py", line 709, in STDERR: main(sys.argv) STDERR: File "DREAM_DM_starter_tf.py", line 704, in main STDERR: train_net(X_tr, X_te, Y_tr, Y_te, opts, f) STDERR: File "DREAM_DM_starter_tf.py", line 601, in train_net STDERR: pred = VGG16_Net(x, 2, keep_prob=keep_prob) STDERR: File "DREAM_DM_starter_tf.py", line 450, in VGG16_Net STDERR: layer = general_conv(X, architecture_conv, b_name=name) STDERR: NameError: global name 'name' is not defined   The command-line for this one was: python DREAM_DM_starter_tf.py \--lr 0.0001 \--reg 0.0005 \--decay 0.985 \--bs 256 \--net VGG16 \--ms 32 \--epoch 10 \--dropout 0.5 \--saver "/modelState/model_vgg16_1.ckpt"   Please fix these bugs. Thanks.   p.s. I might add more items to this post as I continue working with this code.