Testing script

This provides a sanity check to show various 2D and 3D topologies with random data.

python testing.py will test training on a 3D U-Net using randomly-generated, synthetic data.

python testing.py --D2 will test training on a 2D U-Net using randomly-generated, synthetic data.

python testing.py --D2 --inference will test inference on a 2D U-Net using randomly-generated, synthetic data.

usage: testing.py [-h] [--dim_length DIM_LENGTH] [--num_channels NUM_CHANNELS]
                  [--num_outputs NUM_OUTPUTS] [--bz BZ] [--lr LR]
                  [--num_datapoints NUM_DATAPOINTS] [--epochs EPOCHS]
                  [--intraop_threads INTRAOP_THREADS]
                  [--interop_threads INTEROP_THREADS] [--blocktime BLOCKTIME]
                  [--print_model] [--use_upsampling] [--D2]
                  [--single_class_output] [--mkl_verbose] [--inference]
                  [--ngraph] [--keras_api] [--channels_first]

Sanity testing for 3D and 2D Convolution Models

optional arguments:
  -h, --help            show this help message and exit
  --dim_length DIM_LENGTH
                        Tensor cube length of side
  --num_channels NUM_CHANNELS
                        Number of channels
  --num_outputs NUM_OUTPUTS
                        Number of outputs
  --bz BZ               Batch size
  --lr LR               Learning rate
  --num_datapoints NUM_DATAPOINTS
                        Number of datapoints
  --epochs EPOCHS       Number of epochs
  --intraop_threads INTRAOP_THREADS
                        Number of intraop threads
  --interop_threads INTEROP_THREADS
                        Number of interop threads
  --blocktime BLOCKTIME
                        Block time for CPU threads
  --print_model         Print the summary of the model layers
  --use_upsampling      Use upsampling instead of transposed convolution
  --D2                  Use 2D model and images instead of 3D.
  --single_class_output
                        Use binary classifier instead of U-Net
  --mkl_verbose         Print MKL debug statements.
  --inference           Test inference speed. Default=Test training speed
  --ngraph              Use ngraph
  --keras_api           Use Keras API. False=Use tf.keras
  --channels_first      Channels first. NCHW